CN107247926B - A kind of human body detecting method and device - Google Patents
A kind of human body detecting method and device Download PDFInfo
- Publication number
- CN107247926B CN107247926B CN201710359529.6A CN201710359529A CN107247926B CN 107247926 B CN107247926 B CN 107247926B CN 201710359529 A CN201710359529 A CN 201710359529A CN 107247926 B CN107247926 B CN 107247926B
- Authority
- CN
- China
- Prior art keywords
- point cloud
- point
- cloud data
- data
- cluster
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the invention discloses a kind of human body detecting method and devices, to solve parallel overlapping standing, the not high problem of accuracy when carrying out human testing before and after having the even more people of two people in the prior art.This method comprises: being directed to the first point cloud data cluster to be detected, the multiple first sub- point cloud data clusters are divided into according to default rule；Determine the local highest point of each of first point cloud data cluster, if the projector distance of two local highest points in the horizontal plane is less than given threshold, the first sub- point cloud data cluster in the corresponding region of the first sub- point cloud data cluster where this two local highest points is merged into second point cloud aggregate of data, otherwise, it determines the second point cloud aggregate of data comprising the part highest point；According to the HOG value of second point cloud aggregate of data correspondence image, determine whether the image of second point cloud aggregate of data is human body image.The mode of this determining human body can more distinguish parallel the case where being overlapped standing before and after the even more people of two people, improve the accuracy of human testing.
Description
Technical field
The present invention relates to technical field of computer vision, in particular to a kind of human body detecting method and device.
Background technique
Currently, human testing is the research hotspot of computer vision field, have very for the identification and feature extraction of human body
High scientific research and commercial value, has broad application prospects.Traditional human testing is based primarily upon monocular camera acquisition
What the video data that two dimensional image or monocular camera obtain carried out.Due to the single biography in monocular camera or monocular-camera
The ability that sensor obtains information is limited, and characteristics of human body, which extracts, relies primarily on the color and Gradient Features of image, the precision of detection and
Speed is all relatively limited.
It is obtained with appearance of depth camera, such as binocular camera, structure light camera and TOF camera etc., depth camera
Point cloud data can directly obtain the three-dimensional information of specified point in image, and the information of acquisition is more abundant, compensate for traditional list
The deficiency of mesh camera or monocular-camera improves the precision of human testing.
In the prior art, depth camera is when carrying out human testing, first using the consistent (RANdom of random sampling
SAmple Consensus, RANSAC) algorithm determines point cloud data and the deletion of ground level.Secondly using Euclidean distance cluster pair
Remaining point cloud data is clustered, and primarily determines the point cloud data cluster there may be human body.Then true by projection histogram
Vertical Wave crest and wave trough solves the problems, such as one people of more people's one kind and multiclass that cluster generates, and determines that there may be the point cloud numbers of human body again
According to cluster.Finally supported using the training of histograms of oriented gradients (Histograms Of Oriented Gradients, HOG) feature
Vector machine (Support Vector Machine, SVM) classifier carries out human testing.But when environment is complicated, two even
Parallel overlapping is stood before and after more people, is established Wave crest and wave trough by projection histogram and is solved more people's one kind and multiclass one that cluster generates
When the problem of people, which is located in the same projection histogram, so that human body be caused to examine
The accuracy of survey is not high.
Summary of the invention
The present invention provides a kind of human body detecting method and device, to solve in the prior art before having two people even more people
When parallel overlapping is stood afterwards, when carrying out human testing, the not high problem of accuracy.
In order to achieve the above objectives, the embodiment of the invention discloses a kind of human body detecting methods, which comprises
For the first point cloud data cluster to be detected primarily determined as human body, by the first point cloud data cluster according to pre-
If regular partition be the multiple first sub- point cloud data clusters；
According to the highest point in each first sub- point cloud data cluster, the local highest of each of first point cloud data cluster is determined
Point, if the projector distance of two local highest points in the horizontal plane is less than the first threshold of setting, by this two local highests
The first sub- point cloud data cluster in the corresponding region of the sub- point cloud data cluster of the first of point place merges into second point cloud aggregate of data, such as
The projection of fruit part highest point in the horizontal plane is not small at a distance from other any projections of local highest point in the horizontal plane
The second point cloud aggregate of data comprising the part highest point is determined according to the part highest point in the first threshold of setting；
For each second point cloud aggregate of data, the target direction gradient histogram of the second point cloud aggregate of data correspondence image is obtained
Scheme HOG value, according to prediction model trained in advance, determines whether the corresponding image of the second point cloud aggregate of data is human figure
Picture.
Further, it is determined that the process of the first point cloud data cluster includes:
Obtain the point cloud data of image to be detected；
The point cloud data of ground level is determined in the point cloud data of described image using the consistent RANSAC algorithm of random sampling,
The point cloud data of the ground level is deleted in the point cloud data of described image；
It is clustered in the point cloud data after the point cloud data for deleting ground level according to default cluster radius, after cluster
Obtained each classification is determined as the first point cloud data cluster.
Further, described that the first point cloud data cluster is divided into multiple first son point cloud numbers according to default rule
Before cluster, the method also includes:
Identify the highs and lows in each first point cloud data cluster；
If highest point threshold value or minimum point that the highest point in the first point cloud data cluster is higher than setting are most lower than setting
Low spot threshold value then deletes the first point cloud data cluster.
Further, described that the first point cloud data cluster is divided into multiple first son point cloud numbers according to default rule
Include: according to cluster
By the first point cloud data cluster according to preset length and width by the first point cloud data cluster according to default
Regular partition be the multiple first sub- point cloud data clusters, wherein the preset length and width be less than human body cube it is corresponding
Length and width.
Further, the highest point in the sub- point cloud data cluster of the basis each first, determines in the first point cloud data cluster
Each of local highest point include:
For the highest point in each first sub- point cloud data cluster, judge that the highest point in the first sub- point cloud data cluster is
Highest point in no the first sub- point cloud data cluster for preset quantity adjacent thereto, if so, by the first sub- point cloud data
Highest point in cluster is as the local highest point in the first point cloud data cluster.
Further, if the projection of the local highest point in the horizontal plane and other any local highest points are in level
The distance of projection on face is all not less than the first threshold of setting, according to the part highest point, determines to include the part highest point
Second point cloud aggregate of data includes:
According to the part highest point and preset cylinder radius, cylindrical body is determined, will be located at the intracorporal institute of the cylinder
There is point cloud data to be determined as second point cloud aggregate of data.
Further, before the target HOG value for obtaining the second point cloud aggregate of data correspondence image, the method is also wrapped
It includes:
For each second point cloud aggregate of data, the quantity of the point cloud in the second point cloud aggregate of data is identified；
The second point cloud aggregate of data that the quantity of cloud is less than preset second threshold is deleted.
Further, described to be directed to each second point cloud aggregate of data, obtain the mesh of the second point cloud aggregate of data correspondence image
Marking HOG value includes:
For each second point cloud aggregate of data, the highest point in the second point cloud aggregate of data is identified, according to the highest point
And the proportionate relationship of preset height and the width, determine the human region in the corresponding image of the second point cloud aggregate of data；
Obtain the target HOG value of the human region.
The embodiment of the invention discloses a kind of human body detection device, described device includes:
Division module, for being directed to the first point cloud data cluster to be detected primarily determined as human body, by described first point
Cloud aggregate of data is divided into the multiple first sub- point cloud data clusters according to default rule；
Determining module, for determining in the first point cloud data cluster according to the highest point in each first sub- point cloud data cluster
Each of local highest point will if the projector distance of two local highest points in the horizontal plane is less than the first threshold of setting
The first sub- point cloud data cluster in the corresponding region of the first sub- point cloud data cluster where this two local highest points merges into the
Two point cloud data clusters, if the projection of local highest point in the horizontal plane and other any throwings of local highest point in the horizontal plane
The distance of shadow is all not less than the first threshold of setting, according to the part highest point, determines the second point comprising the part highest point
Cloud aggregate of data；
Detection module obtains the mesh of the second point cloud aggregate of data correspondence image for being directed to each second point cloud aggregate of data
Histograms of oriented gradients HOG value is marked, according to prediction model trained in advance, determines the corresponding image of the second point cloud aggregate of data
It whether is human body image.
Further, the determining module is also used to obtain the point cloud data of image to be detected；Using random sampling one
RANSAC algorithm is caused to determine the point cloud data of ground level in the point cloud data of described image, in the point cloud data of described image
Delete the point cloud data of the ground level；According to default cluster radius in the point cloud data after the point cloud data for deleting ground level
It is clustered, each classification obtained after cluster is determined as the first point cloud data cluster.
Further, described device further include:
First identifies removing module, for identification the highs and lows in each first point cloud data cluster；If the
The highest point threshold value or minimum point that highest point in some cloud aggregates of data is higher than setting then will lower than the minimum point threshold value of setting
The first point cloud data cluster is deleted.
Further, the division module is specifically used for the first point cloud data cluster according to preset length and width
The first point cloud data cluster is divided into the multiple first sub- point cloud data clusters according to default rule by degree, wherein described preset
Length and width is less than the corresponding length and width of human body cube.
Further, the determining module, specifically for for the highest point in each first sub- point cloud data cluster, judgement
Highest point in the first sub- point cloud data cluster whether be in the first sub- point cloud data cluster of preset quantity adjacent thereto most
High point, if so, using the highest point in the first sub- point cloud data cluster as the local highest point in the first point cloud data cluster.
Further, the determining module determines circle specifically according to the part highest point and preset cylinder radius
Cylinder will be located at the intracorporal all point cloud datas of the cylinder and be determined as second point cloud aggregate of data.
Further, described device further include:
Second identification removing module identifies in the second point cloud aggregate of data for being directed to each second point cloud aggregate of data
The quantity of point cloud；The second point cloud aggregate of data that the quantity of cloud is less than preset second threshold is deleted.
Further, the detection module is specifically used for being directed to each second point cloud aggregate of data, identifies the second point cloud number
The second point cloud number is determined according to the highest point and the proportionate relationship of preset height and the width according to the highest point in cluster
According to the human region in the corresponding image of cluster；Obtain the target HOG value of the human region.
The embodiment of the invention discloses a kind of human body detecting method and devices, which comprises for be detected first
Step is determined as the first point cloud data cluster of human body, and the first point cloud data cluster is divided into multiple first according to default rule
Sub- point cloud data cluster；According to the highest point in each first sub- point cloud data cluster, each office in the first point cloud data cluster is determined
Portion highest point, if the projector distance of two local highest points in the horizontal plane is less than the first threshold of setting, by two offices
The first sub- point cloud data cluster in the corresponding region of the first sub- point cloud data cluster where portion highest point merges into second point cloud number
According to cluster, if the projection of local highest point in the horizontal plane is at a distance from other any projections of local highest point in the horizontal plane
The second point cloud aggregate of data comprising the part highest point is all determined according to the part highest point not less than the first threshold of setting；
For each second point cloud aggregate of data, the target direction histogram of gradients HOG value of the second point cloud aggregate of data correspondence image is obtained,
According to prediction model trained in advance, determine whether the corresponding image of the second point cloud aggregate of data is human body image.Due to
In the embodiment of the present invention, for the first point cloud data cluster to be detected primarily determined as human body, in the first point cloud data cluster
Local highest point projector distance in the horizontal plane clustered again, determine second point cloud aggregate of data.By obtaining second point
The target HOG value of cloud aggregate of data correspondence image determines whether the corresponding image of the second point cloud aggregate of data is human body image.This
Kind determines the mode of human body, can more distinguish the even more people front and backs of two people and be overlapped the case where standing in parallel, improve human testing
Accuracy.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is a kind of human testing process schematic that the embodiment of the present invention 1 provides；
Fig. 2 is a kind of human testing process schematic that the embodiment of the present invention 3 provides；
Fig. 2A is a kind of first point cloud data cluster schematic diagram provided in an embodiment of the present invention；
Fig. 3 is a kind of human testing process schematic that the embodiment of the present invention 7 provides；
Fig. 4 is a kind of human detection result schematic diagram provided in an embodiment of the present invention；
Fig. 5 is a kind of human body detection device structure chart that the embodiment of the present invention 1 provides；
Fig. 6 is a kind of human body detection device structure chart provided in an embodiment of the present invention.
Specific embodiment
In order to improve the accuracy of the human testing when overlapping is stood in parallel before and after two people even more people, the present invention is implemented
Example provides a kind of human body detecting method and device.
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
Embodiment 1:
Fig. 1 be a kind of human testing process schematic provided in an embodiment of the present invention, the process the following steps are included:
S101: for the first point cloud data cluster to be detected primarily determined as human body, by the first point cloud data cluster
The multiple first sub- point cloud data clusters are divided into according to default rule.
Human body detecting method provided in an embodiment of the present invention can be applied to electronic equipment, and the electronic equipment can be depth
Spend camera.The the first point cloud data cluster to be detected primarily determined as human body is preserved in the electronic equipment, and is preserved
The first point cloud data cluster is divided into the default rule of the multiple first sub- point cloud data clusters.
In embodiments of the present invention, electronic equipment save the first point cloud data cluster in point cloud data based on coordinate
In, it is the plane that is constituted using X-axis and Y-axis as ground level i.e. horizontal plane, Z axis is constructed perpendicular to X/Y plane.For other implementations
The point cloud data in the first point cloud data cluster or other point cloud data clusters in example is also based on the such of this coordinate system.
The first point cloud data cluster is divided into the multiple first sub- point cloud data clusters according to default rule by electronic equipment.
To be detected primarily determine can be one for the first point cloud data cluster of human body, be also possible to it is multiple, each first point
It include multiple point cloud datas in cloud aggregate of data.The first point cloud data cluster is being divided into multiple according to default rule
When one sub- point cloud data cluster, it can be and first determine the corresponding range of all first point cloud data clusters in X-direction and Y direction
Each boundary point determines regional scope according to each boundary point, will in the regional scope according to preset length and width
All first point cloud data clusters are divided into the multiple first sub- point cloud data clusters.Then it determines again in each first point cloud data cluster
The sub- point cloud data cluster of include first.The preset length is the division distance interval of the first point cloud data cluster in X-axis,
The preset width is the division distance interval of the first point cloud data cluster in Y-axis.The length of general human body cube is big
In 30cm, width is greater than 15cm.In view of human body is arbitrarily walked about rotation in space, it is possible that human body and depth camera are not
Parallel situation, the width are also possible to be greater than 30cm.The preset length and width can be identical, be also possible to not
Together, when the length and width is identical, such as can be the length is 10cm, and the width is also 10cm.
S102: according to the highest point in each first sub- point cloud data cluster, each office in the first point cloud data cluster is determined
Portion highest point, if the projector distance of two local highest points in the horizontal plane is less than the first threshold of setting, by two offices
The first sub- point cloud data cluster in the corresponding region of the first sub- point cloud data cluster where portion highest point merges into second point cloud number
According to cluster, if projection and any other local highest point projector distances in the horizontal plane of the local highest point in level are not
The second point cloud aggregate of data comprising the part highest point is determined according to the part highest point less than the first threshold of setting.
It in embodiments of the present invention, include one or more points cloud data, Mei Gedian in each first sub- point cloud data cluster
Cloud data have the information of its corresponding three-dimensional coordinate, and the plane which is constituted using X-axis and Y-axis is i.e. horizontal as ground level
Face, the value on Z axis are the corresponding height value of the point cloud data, identify that the highest point in each first sub- point cloud data cluster exists
It is the point cloud data of maximum value on Z axis, according to the highest point in every sub- point cloud data cluster, determines each first point cloud data
Each of cluster part highest point.Local highest point in each first point cloud data cluster can be one, be also possible to it is multiple,
Local highest point may correspond to the head of human body.
First threshold is preset in electronic equipment, electronic equipment identifies the throwing of two local highest points in the horizontal plane
Shadow distance, and judge whether the distance is less than the first threshold of setting, if it is, thinking this two local highest points pair
The sub- point cloud data cluster of first answered belongs to the point cloud data cluster of the same human body, then by first where this two local highest points
The first sub- point cloud data cluster in the corresponding region of sub- point cloud data cluster merges into second point cloud aggregate of data.
If the projection of the part highest point in the horizontal plane and other any projections of local highest point in the horizontal plane
Distance be all not less than the first threshold of setting, then according to the part highest point, determine include the part highest point second point
Cloud aggregate of data.Each second point cloud aggregate of data is considered a human body.
Highest point in the sub- point cloud data cluster of the basis each first, determines each part in the first point cloud data cluster
When highest point, the specific can be that, by the highest point in each first sub- point cloud data cluster, it is determined as first cloud by comparing
Each of aggregate of data part highest point.Each local highest point is projected to horizontal plane.
If the projection of the part highest point in the horizontal plane and other any projections of local highest point in the horizontal plane
Distance be all not less than the first threshold of setting, it is described according to the part highest point, determine to include the second of the part highest point
Point cloud data cluster, the specific can be that, according to the part highest point and preset length and width, cube is determined, by position
It is determined as second point cloud aggregate of data in described cube of intracorporal all point cloud datas.The part highest point can be located at the cube
Vertex on.As it is assumed that each part highest point is the head of human body, so preferably, which is located at this cube
On the center of the upper surface of body.Each cube intracorporal to have point cloud data to constitute second point cloud aggregate of data to be considered one more
Human body.The length of general human body is greater than 30cm, and the length of human body mentioned here refers to left shoulder edge to the right shoulder side of human body
The distance of edge, the first threshold can be 30cm.
S103: being directed to each second point cloud aggregate of data, obtain the target HOG value of the second point cloud aggregate of data correspondence image,
According to prediction model trained in advance, determine whether the corresponding image of the second point cloud aggregate of data is human body image.
For each second point cloud aggregate of data, the target HOG value of the second point cloud aggregate of data correspondence image is obtained, according to pre-
First trained prediction model, determines whether the corresponding image of the second point cloud aggregate of data is human body image.Obtain this second
When the target HOG value of point cloud data cluster correspondence image, which can be done to YOZ plane or XOZ plane
Two-dimensional projection, the target HOG value of the image after obtaining second point cloud aggregate of data projection.According to prediction mould trained in advance
Type, when determining whether the corresponding image of the second point cloud aggregate of data is human body image, the specific can be that being instructed using HOG feature
Practice SVM classifier and carry out human bioequivalence, whether the image after determining the second point cloud aggregate of data projection is human body image.
Due in embodiments of the present invention, for the first point cloud data cluster to be detected primarily determined as human body, with
The projector distance of local highest point in the horizontal plane in some cloud aggregates of data is clustered again, determines second point cloud aggregate of data.
By obtaining the target HOG value of second point cloud aggregate of data correspondence image, determine that the corresponding image of the second point cloud aggregate of data is
No is human body image.The mode of this determining human body can more distinguish parallel the case where being overlapped standing before and after the even more people of two people, mention
The high accuracy of human testing.
Embodiment 2:
On the basis of the above embodiments, in embodiments of the present invention, the process for determining the first point cloud data cluster includes:
Obtain the point cloud data of image to be detected；
The point cloud data for determining ground level in the point cloud data of described image using RANSAC algorithm, in described image
The point cloud data of the ground level is deleted in point cloud data；
It is clustered in the point cloud data after the point cloud data for deleting ground level according to default cluster radius, after cluster
Obtained each classification is determined as the first point cloud data cluster.
In embodiments of the present invention, electronic equipment obtains the point cloud data of image to be detected, when obtaining described image
Point cloud data after, using RANSAC algorithm in the point cloud data of described image determine ground level point cloud data, and
The point cloud data of the ground level is deleted in the point cloud data of described image.By the left point after the point cloud data for deleting ground level
Cloud data are clustered according to default cluster radius, and each classification obtained after cluster is determined as the first point cloud data cluster.Institute
Stating cluster radius can be related with the concentration of cloud, and point cloud is more intensive, and cluster radius is smaller, and point cloud is more sparse, cluster half
Diameter is bigger.
When electronic equipment is depth camera, electronic equipment is when obtaining the image and point cloud data of scene to be detected, tool
Body can be, and after depth camera is fixed, acquire target scene point cloud data.Depth camera acquisition can be a frame figure
Picture is also possible to one section of video.If it is one section of video, then depth camera can be obtained in this section of video in each frame image
Point cloud data.
When determining the point cloud data of ground level in the point cloud data of described image using RANSAC algorithm, specifically may be used
To be, in the point cloud data of the image, which is considered ground level containing the plane for having the quantity of a cloud most by estimation.
It is also possible to by choosing the point cloud data on arbitrary 3 ground, and the three-dimensional coordinate for including with this 3 point cloud datas manually
Information the equation parameter of 3 point cloud data corresponding flats is calculated by formula AX+BY+CZ+D=0, that is, calculate A, B,
The value of C and D.For all point cloud datas in image, each point cloud data is calculated to European between calculated plane
Distance, then each distance is compared with the distance threshold of setting, if in range corresponding to distance threshold, by the cloud
Data are classified as " intra-office point ", otherwise, are classified as " point not in the know ", are corrected, finally obtain to the ground level being made of " intra-office point "
The point cloud data of all ground levels.The distance threshold can be 15cm.
After the point cloud data of ground level has been determined, the point cloud data of ground level is deleted in the point cloud data of described image
It removes, the point cloud data of deletion is not involved in next each operation.
After deleting the point cloud data of the ground level, by the remaining cloud after the point cloud data for deleting ground level
It is clustered in data according to default cluster radius, each classification obtained after cluster is determined as the first point cloud data cluster.Tool
Body can be, according to preset cluster radius, if between two adjacent point cloud datas in remaining point cloud data
Distance is less than the preset cluster radius, which is determined as same category, is successively traversed all remaining
Point cloud data carries out clustering processing to all remaining point cloud datas, each classification obtained after cluster is determined as at first point
Cloud aggregate of data.The cluster radius is related with the point concentration of cloud, and point cloud data is more intensive, and cluster radius is smaller, puts cloud number
According to more sparse, cluster radius is bigger, and under normal circumstances, which can be 12cm.
When being clustered, any one in point cloud data point cloud can be denoted as pi, will be all in point cloud data
Point cloud is stored in point cloud data cluster Q to be sorted, for the certain point cloud p in Qi, it is drawn off being stored to the C1 subclass in C set
In, identification and piThe distance between be less than cluster radius point cloud pi, by each piIt is stored in C1 subclass from being taken out in Q；So
Another cloud is taken out in remaining cloud of Q again afterwards, is stored in the C2 subclass of C set, above-mentioned cluster half is repeated
The identification of diameter is repeated in the execution above process, until Q is sky, so that each subclass in sorted C set is obtained,
Each subclass is determined as each corresponding type, obtains corresponding each first point cloud data cluster.
Generally, the point cloud data of the point cloud data in the point cloud data in described image to be detected comprising human body, ground level
And other non-human point cloud datas.The point cloud data of ground level is deleted in the point cloud data in described image to be detected,
It is clustered for remaining point cloud data according to default cluster radius, includes human body in the first point cloud data each of after cluster
Point cloud data, but it is also possible that including non-human point cloud data.It may include one in each first point cloud data cluster
Or multiple human bodies.Therefore the first point cloud data cluster after cluster is known as to primarily determine the first point cloud data cluster for human body, it can
To carry out subsequent step for the first point cloud data cluster, the second point cloud data comprising human body is determined more accurately out
Cluster, and then detect human body.
Embodiment 3:
In order to avoid wall, the interference of the too high or too low object such as desk further increases the accuracy of human testing,
On the basis of the various embodiments described above, in embodiments of the present invention, it is described by the first point cloud data cluster according to preset rule
It is then divided into before the multiple first sub- point cloud data clusters, the method also includes:
Identify the highs and lows in each first point cloud data cluster；
If highest point threshold value or minimum point that the highest point in the first point cloud data cluster is higher than setting are most lower than setting
Low spot threshold value then deletes the first point cloud data cluster.
In embodiments of the present invention, highest point threshold value and minimum point threshold value are preserved in electronic equipment.For to be detected
It primarily determines the first point cloud data cluster for human body, identifies the highs and lows in each first point cloud data cluster, judgement
Whether the highest point in each first point cloud data cluster is greater than the highest point threshold value of setting, if it is, the highest point is corresponding
The first point cloud data cluster delete.Judge whether the minimum point in each first point cloud data cluster is less than the minimum point threshold of setting
Value, if it is, the corresponding first point cloud data cluster of the minimum point is deleted.
The the first point cloud data cluster deleted is not involved in next each operation.The highest point threshold value of the setting can be
Bigger numerical value, such as can be 2.3 meters, 2.5 meters etc., the minimum point threshold value of setting can be smaller numerical value, such as
It can be 1 meter, 1.3 meters etc., since it typically be considered that the height of human body is no more than highest point threshold value, and be not less than minimum point threshold
Value.If the highest point in the first point cloud data cluster is higher than highest point threshold value, such as 2.3 meters, then it is assumed that first cloud number
It may be metope according to cluster, so being deleted.If the minimum point in the first point cloud data cluster is lower than minimum point threshold value, such as
1.3 meters, then it is assumed that the first point cloud data cluster may be the first non-human point cloud data clusters such as desk, chair, so by it
It deletes.
In order to improve the accuracy of human testing, for the first point cloud data cluster to be detected primarily determined as human body,
It is also possible to delete the point cloud data of the highest point threshold value for being higher than setting in each first point cloud data cluster, and lower than setting
Minimum point threshold value point cloud data delete.The point cloud data of deletion is not involved in next each operation.
Fig. 2 is a kind of human testing process schematic that provides of the embodiment of the present invention 3, the process the following steps are included:
S201: the point cloud data of image to be detected is obtained.
S202: the point cloud data of ground level is determined in the point cloud data of described image using RANSAC algorithm, described
The point cloud data of the ground level is deleted in the point cloud data of image.
S203: being clustered in the point cloud data after the point cloud data for deleting ground level according to default cluster radius, will
The each classification obtained after cluster is determined as the first point cloud data cluster.
S204: for the first point cloud data cluster to be detected primarily determined as human body, each first point cloud data is identified
Highs and lows in cluster, if the highest point in the first point cloud data cluster is higher than the highest point threshold value or minimum of setting
Point then deletes the first point cloud data cluster lower than the minimum point threshold value of setting.
S205: the remaining first point cloud data cluster is divided into the multiple first sub- point cloud datas according to default rule
Cluster.
S206: according to the highest point in each first sub- point cloud data cluster, each office in the first point cloud data cluster is determined
Portion highest point, if the projector distance of two local highest points in the horizontal plane is less than the first threshold of setting, by two offices
The first sub- point cloud data cluster in the corresponding region of the first sub- point cloud data cluster where portion highest point merges into second point cloud number
According to cluster, if the projection of local highest point in the horizontal plane is at a distance from other any projections of local highest point in the horizontal plane
The second point cloud aggregate of data comprising the part highest point is all determined according to the part highest point not less than the first threshold of setting.
S207: being directed to each second point cloud aggregate of data, obtains the target direction ladder of the second point cloud aggregate of data correspondence image
Histogram HOG value is spent, according to prediction model trained in advance, determines whether the corresponding image of the second point cloud aggregate of data is people
Body image.
Embodiment 4:
In order to further increase the accuracy of human testing, it is described by the first point cloud data cluster according to default rule
Being divided into the multiple first sub- point cloud data clusters includes:
By the first point cloud data cluster according to preset length and width by the first point cloud data cluster according to default
Regular partition be the multiple first sub- point cloud data clusters, wherein the preset length and width be less than human body cube it is corresponding
Length and width.
In embodiments of the present invention, the first point cloud data cluster is being divided into multiple first sons according to default rule
When point cloud data cluster, the specific can be that, by the first point cloud data cluster according to preset length and width by described first
Point cloud data cluster is divided into the multiple first sub- point cloud data clusters, wherein the preset length and width is less than human body cube pair
The length and width answered.The preset length is the division distance interval of the first point cloud data cluster in X-axis, described default
Width be the division distance interval of the first point cloud data cluster in Y-axis.The length of general human body cube is greater than 30cm, wide
Degree is greater than 15cm.The preset length and width can be identical, be also possible to difference, when the length and width is identical
When, such as can be the length is 10cm, the width is also 10cm.
Embodiment 5:
The each part by the highest point in each first sub- point cloud data cluster, being determined as in the first point cloud data cluster most
When high point, the inaccuracy of subsequent human testing may cause, in order to further increase the accuracy of human testing.In above-mentioned each reality
On the basis of applying example, in embodiments of the present invention, the highest point in the sub- point cloud data cluster of the basis each first determines first
Each of point cloud data cluster part highest point includes:
For the highest point in each first sub- point cloud data cluster, judge that the highest point in the first sub- point cloud data cluster is
Highest point in no the first sub- point cloud data cluster for setting quantity adjacent thereto, if so, by the first sub- point cloud data
Highest point in cluster is as the local highest point in the first point cloud data cluster.
In embodiments of the present invention, for each point cloud data cluster, the multiple first sub- points are divided into according to default rule
Cloud aggregate of data, the position of each first sub- point cloud data cluster be clearly, therefore be clear that and this first son point cloud
Each of the adjacent first sub- point cloud data cluster of aggregate of data.According to the highest point in each first sub- point cloud data cluster, determine
When the highest point of each of first point cloud data cluster part, the specific can be that, in each first sub- point cloud data cluster
Highest point, judge the highest point in the first sub- point cloud data cluster whether be preset quantity adjacent thereto first son point cloud number
According to the highest point in cluster, if so, using the highest point in the first sub- point cloud data cluster as the office in the first point cloud data cluster
Portion highest point.The preset quantity can be 4, or 8.
Fig. 2A is a kind of first point cloud data cluster schematic diagram provided in an embodiment of the present invention, it can be seen from Fig. 2A this
Some cloud aggregates of data include 12 the first sub- point cloud data clusters, mark in the first point cloud data cluster in the perspective view of XOY plane
Number be 1-12.For the 6th the first sub- point cloud data cluster, the first son point cloud adjacent with the 6th the first sub- point cloud data cluster
Aggregate of data has 8, and label is respectively 1,2,3,5,7,9,10,11.When the preset quantity is 8, then the 6th the is judged
Whether highest point in one sub- point cloud data cluster is highest point in the sub- point cloud data cluster of all first adjacent thereto, if
It is, using the highest point in the 6th the first sub- point cloud data cluster as the local highest point in the first point cloud data cluster.If institute
Stating preset quantity is 4, then judges whether the highest point in the 6th the first sub- point cloud data cluster is any 4 adjacent thereto
Highest point in first sub- point cloud data cluster, if so, using the highest point in the 6th the first sub- point cloud data cluster as first
Local highest point in point cloud data cluster.Selecting the 4 first son point clouds adjacent with the 6th first sub- point cloud data cluster
When aggregate of data, it can be and arbitrarily select 4, preferably, selection and the 6th the first sub- point in 1,2,3,5,7,9,10,11
The sub- point cloud data cluster of the first of the adjacent upper and lower, left and right of cloud aggregate of data, i.e. the first sub- point cloud data marked as 2,5,7,10
Cluster.
Local highest point in each first point cloud data cluster can be one, be also possible to multiple, each part highest
Point may correspond to the head of human body.
Embodiment 6:
On the basis of the various embodiments described above, in embodiments of the present invention, if the local highest point is in the horizontal plane
Projection be all not less than the first threshold of setting at a distance from other any projections of local highest point in the horizontal plane, according to this
Local highest point determines to include that the part highest point second point cloud aggregate of data includes:
According to the part highest point and preset cylinder radius, cylindrical body is determined, will be located at the intracorporal institute of the cylinder
There is point cloud data to be determined as second point cloud aggregate of data.
In embodiments of the present invention, according to the part highest point, the second point cloud number comprising the part highest point is determined
When according to cluster, the specific can be that determining cylindrical body according to the part highest point and preset cylinder radius, the circle will be located at
The intracorporal all point cloud datas of column are determined as second point cloud aggregate of data.The diameter of the cylindrical body can be identical as first threshold,
It can also be different, when the diameter of cylindrical body is identical as first threshold, the first threshold is described preset if it is 30cm
Cylinder radius is 15cm.The preset cylinder diameter is greater than preset cluster radius.
Embodiment 7:
In order to further increase the accuracy of human testing, on the basis of the various embodiments described above, in the present invention is implemented,
Before the target HOG value for obtaining the second point cloud aggregate of data correspondence image, the method also includes:
For each second point cloud aggregate of data, the quantity of the point cloud in the second point cloud aggregate of data is identified；
The second point cloud aggregate of data that the quantity of cloud is less than preset second threshold is deleted.
In embodiments of the present invention, preset second threshold is preserved in electronic equipment.When the second point cloud data has been determined
After cluster, it can be directed to each second point cloud aggregate of data, the quantity of the point cloud in the second point cloud aggregate of data is identified, judge the point
Whether the quantity of cloud is less than default second threshold, if it is, the quantity of cloud to be less than to the second point of preset second threshold
Cloud aggregate of data is deleted.
When the point cloud quantity in the second point cloud aggregate of data is less than preset second threshold, people may be corresponded to and lifted
The corresponding second point cloud aggregate of data of the arm is then deleted, has no effect on the detection to human body by arm.
Fig. 3 is a kind of human testing process schematic that provides of the embodiment of the present invention 7, the process the following steps are included:
S301: for the first point cloud data cluster to be detected primarily determined as human body, each first point cloud data is identified
Highs and lows in cluster, if the highest point in the first point cloud data cluster is higher than the highest point threshold value or minimum of setting
Point then deletes the first point cloud data cluster lower than the minimum point threshold value of setting.
S302: the remaining first point cloud data cluster is divided into the multiple first sub- point cloud datas according to default rule
Cluster.
S303: according to the highest point in each first sub- point cloud data cluster, each office in the first point cloud data cluster is determined
Portion highest point, if the projector distance of two local highest points in the horizontal plane is less than the first threshold of setting, by two offices
The first sub- point cloud data cluster in the corresponding region of the first sub- point cloud data cluster where portion highest point merges into second point cloud number
According to cluster, if the projection of local highest point in the horizontal plane is at a distance from other any projections of local highest point in the horizontal plane
The second point cloud aggregate of data comprising the part highest point is all determined according to the part highest point not less than the first threshold of setting.
S304: being directed to each second point cloud aggregate of data, identifies the quantity of the point cloud in the second point cloud aggregate of data, will put cloud
Quantity be less than preset second threshold second point cloud aggregate of data delete.
S305: it is directed to remaining each second point cloud aggregate of data, obtains the target of the second point cloud aggregate of data correspondence image
Histograms of oriented gradients HOG value determines that the corresponding image of the second point cloud aggregate of data is according to prediction model trained in advance
No is human body image.
Embodiment 8:
In order to further increase the accuracy of human testing, on the basis of the various embodiments described above, in the present invention is implemented,
Described to be directed to each second point cloud aggregate of data, the target HOG value for obtaining the second point cloud aggregate of data correspondence image includes:
For each second point cloud aggregate of data, the highest point in the second point cloud aggregate of data is identified, according to the highest point
And the proportionate relationship of preset height and the width, determine the human region in the corresponding image of the second point cloud aggregate of data；
Obtain the target HOG value of the human region.
In embodiments of the present invention, after each second point cloud aggregate of data has been determined, if X-axis is put down with what Y-axis was constituted
Face is ground level, which can be done two-dimensional projection to YOZ plane or XOZ plane, after identification projection
Highest point in second point cloud aggregate of data, according to the highest point and the proportionate relationship of preset height and the width, determine described in
Human region in the corresponding image of second point cloud aggregate of data obtains the target HOG value of the human region.Each highest point
It is considered the head of people, highest point is higher, then corresponding human region is bigger.
When determining the human region in the corresponding image of the second point cloud aggregate of data, it is every to be specifically also possible to identification
Highest point in a second point cloud aggregate of data determines square according to the highest point and the proportionate relationship of preset height and the width
The rectangular area is extended preset length to depth camera direction, forms 3-D image by shape region, by the 3-D image to
YOZ plane or XOZ plane do two-dimensional projection, and the two-dimensional projection is determined as the corresponding image of the second point cloud aggregate of data
In human region.The preset length is 10cm.
Fig. 4 is a kind of human detection result schematic diagram provided in an embodiment of the present invention, and each of Fig. 4 body is all by it
Corresponding annulus entangles, and illustrates that the human body is detected.
Fig. 5 is a kind of human body detection device structure chart that the embodiment of the present invention 1 provides, and described device includes:
Division module 51, for being directed to the first point cloud data cluster to be detected primarily determined as human body, by described first
Point cloud data cluster is divided into the multiple first sub- point cloud data clusters according to default rule；
Determining module 52, for determining the first point cloud data cluster according to the highest point in each first sub- point cloud data cluster
Each of local highest point, if the projector distance of two local highest points in the horizontal plane is less than the first threshold of setting,
The first sub- point cloud data cluster in the corresponding region of the first sub- point cloud data cluster where this two local highest points is merged into
Second point cloud aggregate of data, if the projection of local highest point in the horizontal plane and other any local highest points are in the horizontal plane
The distance of projection is all not less than the first threshold of setting, according to the part highest point, determines to include the second of the part highest point
Point cloud data cluster；
Detection module 53 obtains the second point cloud aggregate of data correspondence image for being directed to each second point cloud aggregate of data
Target direction HOG value determines whether the corresponding image of the second point cloud aggregate of data is people according to prediction model trained in advance
Body image.
The determining module 52, is also used to obtain the point cloud data of image to be detected；Using RANSAC algorithm described
The point cloud data that ground level is determined in the point cloud data of image, deletes the point of the ground level in the point cloud data of described image
Cloud data；It is clustered in the point cloud data after the point cloud data for deleting ground level according to default cluster radius, after cluster
Obtained each classification is determined as the first point cloud data cluster.
Fig. 6 is a kind of human body detection device structure chart provided in an embodiment of the present invention, on the basis of Fig. 5, described device
Further include:
First identifies removing module 61, for identification the highs and lows in each first point cloud data cluster；If
The highest point threshold value or minimum point that highest point in first point cloud data cluster is higher than setting are lower than the minimum point threshold value set, then
The first point cloud data cluster is deleted.
The division module 51, being specifically used for will be described according to preset length and width by the first point cloud data cluster
First point cloud data cluster is divided into the multiple first sub- point cloud data clusters according to default rule, wherein the preset length and width
Degree is less than the corresponding length and width of human body cube.
The determining module 52, specifically for for the highest point in each first sub- point cloud data cluster, judge this first
Whether highest point in sub- point cloud data cluster is highest point in the first sub- point cloud data cluster of preset quantity adjacent thereto, such as
Fruit is, using the highest point in the first sub- point cloud data cluster as the local highest point in the first point cloud data cluster.
The determining module 52 determines cylindrical body specifically according to the part highest point and preset cylinder radius, will
It is determined as second point cloud aggregate of data positioned at the intracorporal all point cloud datas of the cylinder.
Described device further include:
Second identification removing module 62 is also used to identify the second point cloud aggregate of data for each second point cloud aggregate of data
In point cloud quantity；The second point cloud aggregate of data that the quantity of cloud is less than preset second threshold is deleted.
The detection module 53 is specifically used for being directed to each second point cloud aggregate of data, identify in the second point cloud aggregate of data
Highest point the second point cloud aggregate of data pair is determined according to the highest point and the proportionate relationship of preset height and the width
The human region in image answered；Obtain the target HOG value of the human region.
The embodiment of the invention discloses a kind of human body detecting method and devices, which comprises for be detected first
Step is determined as the first point cloud data cluster of human body, and the first point cloud data cluster is divided into multiple first according to default rule
Sub- point cloud data cluster；According to the highest point in each first sub- point cloud data cluster, each office in the first point cloud data cluster is determined
Portion highest point, if the projector distance of two local highest points in the horizontal plane is less than the first threshold of setting, by two offices
The first sub- point cloud data cluster in the corresponding region of the first sub- point cloud data cluster where portion highest point merges into second point cloud number
According to cluster, if the projection of local highest point in the horizontal plane is at a distance from other any projections of local highest point in the horizontal plane
The second point cloud aggregate of data comprising the part highest point is all determined according to the part highest point not less than the first threshold of setting；
For each second point cloud aggregate of data, the target direction histogram of gradients HOG value of the second point cloud aggregate of data correspondence image is obtained,
According to prediction model trained in advance, determine whether the corresponding image of the second point cloud aggregate of data is human body image.Due to
In the embodiment of the present invention, for the first point cloud data cluster to be detected primarily determined as human body, in the first point cloud data cluster
Local highest point projector distance in the horizontal plane clustered again, determine second point cloud aggregate of data.By obtaining second point
The target HOG value of cloud aggregate of data correspondence image determines whether the corresponding image of the second point cloud aggregate of data is human body image.This
Kind determines the mode of human body, can more distinguish the even more people front and backs of two people and be overlapped the case where standing in parallel, improve human testing
Accuracy.
For systems/devices embodiment, since it is substantially similar to the method embodiment, so the comparison of description is simple
Single, the relevent part can refer to the partial explaination of embodiments of method.
It should be noted that, in this document, relational terms such as first and second and the like are used merely to a reality
Body or an operation are distinguished with another entity or another operation, without necessarily requiring or implying these entities
Or there are any actual relationship or orders between operation.
It should be understood by those skilled in the art that, embodiments herein can provide as method, system or computer program
Product.Therefore, the reality of complete hardware embodiment, complete Application Example or connected applications and hardware aspect can be used in the application
Apply the form of example.Moreover, it wherein includes the computer of computer usable program code that the application, which can be used in one or more,
The computer program implemented in usable storage medium (including but not limited to magnetic disk storage, CD-ROM, optical memory etc.) produces
The form of product.
The application is referring to method, the process of equipment (system) and computer program product according to the embodiment of the present application
Figure and/or block diagram describe.It should be understood that every one stream in flowchart and/or the block diagram can be realized by computer program instructions
The combination of process and/or box in journey and/or box and flowchart and/or the block diagram.It can provide these computer programs
Instruct the processor of general purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices to produce
A raw machine, so that being generated by the instruction that computer or the processor of other programmable data processing devices execute for real
The device for the function of being specified in present one or more flows of the flowchart and/or one or more blocks of the block diagram.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable data processing devices with spy
Determine in the computer-readable memory that mode works, so that it includes referring to that instruction stored in the computer readable memory, which generates,
Enable the manufacture of device, the command device realize in one box of one or more flows of the flowchart and/or block diagram or
The function of being specified in multiple boxes.
These computer program instructions also can be loaded onto a computer or other programmable data processing device, so that counting
Series of operation steps are executed on calculation machine or other programmable devices to generate computer implemented processing, thus in computer or
The instruction executed on other programmable devices is provided for realizing in one or more flows of the flowchart and/or block diagram one
The step of function of being specified in a box or multiple boxes.
Although the preferred embodiment of the application has been described, it is created once a person skilled in the art knows basic
Property concept, then additional changes and modifications can be made to these embodiments.So it includes excellent that the following claims are intended to be interpreted as
It selects embodiment and falls into all change and modification of the application range.
Obviously, those skilled in the art can carry out various modification and variations without departing from the essence of the application to the application
Mind and range.In this way, if these modifications and variations of the application belong to the range of the claim of this application and its equivalent technologies
Within, then the application is also intended to include these modifications and variations.
Claims (16)
1. a kind of human body detecting method, which is characterized in that the described method includes:
For the first point cloud data cluster to be detected primarily determined as human body, by the first point cloud data cluster according to preset
Regular partition is the multiple first sub- point cloud data clusters；
According to the highest point in each first sub- point cloud data cluster, the local highest point of each of first point cloud data cluster is determined,
If the projector distance of two local highest points in the horizontal plane is less than the first threshold of setting, by this two local highest point institutes
The corresponding region of the first sub- point cloud data cluster in the first sub- point cloud data cluster merge into second point cloud aggregate of data, if office
The projection of portion highest point in the horizontal plane is all not less than at a distance from other any projections of local highest point in the horizontal plane to be set
Fixed first threshold determines the second point cloud aggregate of data comprising the part highest point according to the part highest point；
For each second point cloud aggregate of data, the target direction histogram of gradients of the second point cloud aggregate of data correspondence image is obtained
HOG value determines whether the corresponding image of the second point cloud aggregate of data is human body image according to prediction model trained in advance.
2. the method as described in claim 1, which is characterized in that the process for determining the first point cloud data cluster includes:
Obtain the point cloud data of image to be detected；
The point cloud data for determining ground level in the point cloud data of described image using the consistent RANSAC algorithm of random sampling, in institute
State the point cloud data that the ground level is deleted in the point cloud data of image；
It is clustered, will be obtained after cluster according to default cluster radius in the point cloud data after the point cloud data for deleting ground level
Each classification be determined as the first point cloud data cluster.
3. the method as described in claim 1, which is characterized in that it is described by the first point cloud data cluster according to default rule
It is divided into before the multiple first sub- point cloud data clusters, the method also includes:
Identify the highs and lows in each first point cloud data cluster；
If highest point threshold value or minimum point that the highest point in the first point cloud data cluster is higher than setting are lower than the minimum point set
Threshold value then deletes the first point cloud data cluster.
4. the method as described in claim 1, which is characterized in that it is described by the first point cloud data cluster according to default rule
Being divided into the multiple first sub- point cloud data clusters includes:
By the first point cloud data cluster according to preset length and width by the first point cloud data cluster according to preset rule
The multiple first sub- point cloud data clusters are then divided into, wherein the preset length and width is less than the corresponding length of human body cube
And width.
5. the method as described in claim 1, which is characterized in that the highest in the sub- point cloud data cluster of the basis each first
Point determines that the local highest point of each of first point cloud data cluster includes:
For the highest point in each first sub- point cloud data cluster, judge the highest point in the first sub- point cloud data cluster whether be
Highest point in the sub- point cloud data cluster of the first of preset quantity adjacent thereto, if so, by the first sub- point cloud data cluster
Highest point as the local highest point in the first point cloud data cluster.
6. the method as described in claim 1, which is characterized in that if the projection of the local highest point in the horizontal plane with appoint
The distance of other projections of local highest point in the horizontal plane of anticipating all is not less than the first threshold of setting, according to the part highest
Point determines to include that the part highest point second point cloud aggregate of data includes:
According to the part highest point and preset cylinder radius, cylindrical body is determined, will be located at the intracorporal all the points of the cylinder
Cloud data are determined as second point cloud aggregate of data.
7. method as claimed in claim 1 or 3, which is characterized in that the acquisition second point cloud aggregate of data correspondence image
Before target HOG value, the method also includes:
For each second point cloud aggregate of data, the quantity of the point cloud in the second point cloud aggregate of data is identified；
The second point cloud aggregate of data that the quantity of cloud is less than preset second threshold is deleted.
8. the method as described in claim 1, which is characterized in that it is described be directed to each second point cloud aggregate of data, obtain this second
The target HOG value of point cloud data cluster correspondence image includes:
For each second point cloud aggregate of data, the highest point in the second point cloud aggregate of data is identified, according to the highest point and in advance
If height and the width proportionate relationship, determine the human region in the corresponding image of the second point cloud aggregate of data；
Obtain the target HOG value of the human region.
9. a kind of human body detection device, which is characterized in that described device includes:
Division module, for being directed to the first point cloud data cluster to be detected primarily determined as human body, by first cloud number
The multiple first sub- point cloud data clusters are divided into according to default rule according to cluster；
Determining module, for determining every in the first point cloud data cluster according to the highest point in each first sub- point cloud data cluster
A part highest point, if the projector distance of two local highest points in the horizontal plane is less than the first threshold of setting, by this two
The first sub- point cloud data cluster in the corresponding region of the first sub- point cloud data cluster where a part highest point merges into second point
Cloud aggregate of data, if the projection of local highest point in the horizontal plane and other any projections of local highest point in the horizontal plane
Distance all determines the second point cloud number comprising the part highest point according to the part highest point not less than the first threshold of setting
According to cluster；
Detection module obtains the target side of the second point cloud aggregate of data correspondence image for being directed to each second point cloud aggregate of data
Whether the corresponding image of the second point cloud aggregate of data is determined according to prediction model trained in advance to histogram of gradients HOG value
For human body image.
10. device as claimed in claim 9, which is characterized in that the determining module is also used to obtain image to be detected
Point cloud data；The point cloud number of ground level is determined in the point cloud data of described image using the consistent RANSAC algorithm of random sampling
According to deleting the point cloud data of the ground level in the point cloud data of described image；After the point cloud data for deleting ground level
It is clustered in point cloud data according to default cluster radius, each classification obtained after cluster is determined as the first point cloud data
Cluster.
11. device as claimed in claim 9, which is characterized in that described device further include:
First identifies removing module, for identification the highs and lows in each first point cloud data cluster；If first point
Highest point in cloud aggregate of data be higher than setting highest point threshold value or minimum point lower than setting minimum point threshold value, then by this
Some cloud aggregates of data are deleted.
12. device as claimed in claim 9, which is characterized in that the division module is specifically used for first cloud number
The first point cloud data cluster is divided into the multiple first sub- points according to default rule according to preset length and width according to cluster
Cloud aggregate of data, wherein the preset length and width is less than the corresponding length and width of human body cube.
13. device as claimed in claim 9, which is characterized in that the determining module is specifically used for being directed to each first sub- point
Highest point in cloud aggregate of data judges whether the highest point in the first sub- point cloud data cluster is preset quantity adjacent thereto
Highest point in first sub- point cloud data cluster, if so, using the highest point in the first sub- point cloud data cluster as first cloud
Local highest point in aggregate of data.
14. device as claimed in claim 9, which is characterized in that the determining module, specifically according to the part highest point and
Preset cylinder radius, determines cylindrical body, will be located at the intracorporal all point cloud datas of the cylinder and be determined as second point cloud number
According to cluster.
15. the device as described in claim 9 or 11, which is characterized in that described device further include:
Second identification removing module identifies the point cloud in the second point cloud aggregate of data for being directed to each second point cloud aggregate of data
Quantity；The second point cloud aggregate of data that the quantity of cloud is less than preset second threshold is deleted.
16. device as claimed in claim 9, which is characterized in that the detection module is specifically used for being directed to each second point cloud
Aggregate of data identifies the highest point in the second point cloud aggregate of data, according to the highest point and the ratio of preset height and the width
Relationship determines the human region in the corresponding image of the second point cloud aggregate of data；Obtain the target HOG of the human region
Value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710359529.6A CN107247926B (en) | 2017-05-19 | 2017-05-19 | A kind of human body detecting method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710359529.6A CN107247926B (en) | 2017-05-19 | 2017-05-19 | A kind of human body detecting method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107247926A CN107247926A (en) | 2017-10-13 |
| CN107247926B true CN107247926B (en) | 2019-09-17 |
Family
ID=60017347
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710359529.6A Active CN107247926B (en) | 2017-05-19 | 2017-05-19 | A kind of human body detecting method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107247926B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109214348A (en) * | 2018-09-19 | 2019-01-15 | 北京极智嘉科技有限公司 | A kind of obstacle detection method, device, equipment and storage medium |
| CN109740628A (en) * | 2018-12-03 | 2019-05-10 | 深圳市华讯方舟太赫兹科技有限公司 | Point cloud clustering method, image processing equipment and the device with store function |
| TWI804671B (en) | 2019-08-28 | 2023-06-11 | 財團法人工業技術研究院 | Interaction display method and interaction display system |
| CN112651266A (en) * | 2019-10-11 | 2021-04-13 | 北京百度网讯科技有限公司 | Pedestrian detection method and device |
| CN112070052A (en) * | 2020-09-16 | 2020-12-11 | 青岛维感科技有限公司 | Interval monitoring method, device and system and storage medium |
| CN112183381A (en) * | 2020-09-30 | 2021-01-05 | 深兰人工智能(深圳)有限公司 | Method and device for detecting driving area of vehicle |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2562688A2 (en) * | 2011-08-22 | 2013-02-27 | Samsung Electronics Co., Ltd. | Method of Separating Object in Three Dimensional Point Cloud |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102930246B (en) * | 2012-10-16 | 2015-04-08 | 同济大学 | Indoor scene identifying method based on point cloud fragment division |
| CN104298998B (en) * | 2014-09-28 | 2017-09-08 | 北京理工大学 | A kind of data processing method of 3D point cloud |
| CN104408445A (en) * | 2014-12-19 | 2015-03-11 | 吉林纪元时空动漫游戏科技股份有限公司 | Automatic real-time human body detecting method |
| CN104751146B (en) * | 2015-04-13 | 2018-04-10 | 中国科学技术大学 | A kind of indoor human body detection method based on 3D point cloud image |
| CN105674991A (en) * | 2016-03-29 | 2016-06-15 | 深圳市华讯方舟科技有限公司 | Robot positioning method and device |
-
2017
- 2017-05-19 CN CN201710359529.6A patent/CN107247926B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2562688A2 (en) * | 2011-08-22 | 2013-02-27 | Samsung Electronics Co., Ltd. | Method of Separating Object in Three Dimensional Point Cloud |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107247926A (en) | 2017-10-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107247926B (en) | A kind of human body detecting method and device | |
| CN106650640A (en) | Negative obstacle detection method based on local structure feature of laser radar point cloud | |
| CN106647742B (en) | Movement routine method and device for planning | |
| EP3570253B1 (en) | Method and device for reconstructing three-dimensional point cloud | |
| CN105068918B (en) | A kind of page method of testing and device | |
| GB2532948A (en) | Objection recognition in a 3D scene | |
| CN105205866A (en) | Dense-point-cloud-based rapid construction method of urban three-dimensional model | |
| CN109035330A (en) | Cabinet approximating method, equipment and computer readable storage medium | |
| GB2543749A (en) | 3D scene rendering | |
| CN104050709B (en) | A kind of three dimensional image processing method and electronic equipment | |
| CN103247045A (en) | Method of obtaining artificial scene main directions and image edges from multiple views | |
| JP7156515B2 (en) | Point cloud annotation device, method and program | |
| CN104657968B (en) | Automatic vehicle-mounted three-dimensional laser point cloud facade classification and outline extraction method | |
| US20150138193A1 (en) | Method and device for panorama-based inter-viewpoint walkthrough, and machine readable medium | |
| Pound et al. | A patch-based approach to 3D plant shoot phenotyping | |
| CN108225334A (en) | A kind of localization method and device based on three-dimensional live-action data | |
| CN106022266A (en) | Target tracking method and target tracking apparatus | |
| CN108665472A (en) | The method and apparatus of point cloud segmentation | |
| CN106327576B (en) | A kind of City scenarios method for reconstructing and system | |
| CN110084797A (en) | Plane monitoring-network method, apparatus, electronic equipment and storage medium | |
| Özdemir et al. | A multi-purpose benchmark for photogrammetric urban 3D reconstruction in a controlled environment | |
| Bobkov et al. | Room segmentation in 3D point clouds using anisotropic potential fields | |
| CN111870952A (en) | Height map generation method, device, equipment and storage medium | |
| KR20110094957A (en) | Apparatus and method for object segmentation from range image | |
| CN107452012A (en) | A kind of pedestrian recognition method, device and electronic equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |