CN110245564A - A kind of pedestrian detection method, system and terminal device - Google Patents
A kind of pedestrian detection method, system and terminal device Download PDFInfo
- Publication number
- CN110245564A CN110245564A CN201910397924.2A CN201910397924A CN110245564A CN 110245564 A CN110245564 A CN 110245564A CN 201910397924 A CN201910397924 A CN 201910397924A CN 110245564 A CN110245564 A CN 110245564A
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- target image
- sample
- positive sample
- scene
- 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.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 53
- 238000012549 training Methods 0.000 claims abstract description 30
- 238000000034 method Methods 0.000 claims abstract description 19
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 18
- 238000012544 monitoring process Methods 0.000 claims abstract description 11
- 238000004590 computer program Methods 0.000 claims description 12
- 238000013480 data collection Methods 0.000 claims description 11
- 238000010586 diagram Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 6
- 238000013135 deep learning Methods 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 230000006399 behavior Effects 0.000 description 2
- 238000004140 cleaning Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 238000009472 formulation Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (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 present invention is suitable for image identification technical field, provides a kind of pedestrian detection method, system and terminal device, and method includes: by the multipair pedestrian as in convolution depth network model recognition target image, and in the target image to pedestrian's addition the first identification frame；Wherein, the multipair training mission as convolution depth network model includes semantic task and pedestrian detection；By the certain objects in convolutional neural networks VGG19 recognition target image, and in the target image to certain objects addition the second identification frame；It is non-overlapping to judge that the first identification frame and the second identification frame have, if there is overlapping, determines that pedestrian carries certain objects, and cause preset monitor event.Accurate pedestrian detection, and the monitoring of various pedestrian's attribute, scene properties may be implemented through the invention.
Description
Technical field
The present invention relates to image identification technical field more particularly to a kind of pedestrian detection methods, system and terminal device.
Background technique
Pedestrian detection (Pedestrian Detection) is always the hot and difficult issue in computer vision research, most
In recent years by very big concern.Pedestrian detection will solve the problems, such as: find out pedestrian all in image or video frame, including position
It sets and size, is generally indicated with rectangle frame, this is typical target detection problems and general way.
However, in public safety, monitoring for public place pedestrian not only needs to detect pedestrian, label
The position of pedestrian out, more needs to be concerned with and avoids accidentally identifying, and not be by other objects with humanoid profile are mis-marked
People.In addition, the important belongings for pedestrian are also paid close attention to very much, the especially monitoring of specific behavior has realistic meaning very much,
For example it needs to pay special attention to for carrying the pedestrian of package in important place.
Summary of the invention
It is a primary object of the present invention to propose a kind of pedestrian detection method, system and terminal device, to solve existing skill
When being monitored in art to public place pedestrian, pedestrian detection error is big, and does not have the problem of special article detection function.
To achieve the above object, first aspect of the embodiment of the present invention provides a kind of pedestrian detection method, comprising:
By the multipair pedestrian as in convolution depth network model recognition target image, and to institute in the target image
State pedestrian's addition the first identification frame；
Wherein, the multipair training mission as convolution depth network model includes semantic task and pedestrian detection；
The certain objects in the target image are identified by convolutional neural networks VGG19, and in the target image
To certain objects addition the second identification frame；
It is non-overlapping to judge that the first identification frame and the second identification frame have, if there is overlapping, determines that the pedestrian takes
With the certain objects, and cause preset monitor event.
In conjunction with the embodiment of the present invention in a first aspect, by multipair as convolution described in first embodiment of the embodiment of the present invention
Pedestrian in depth network model recognition target image, comprising:
Obtain pedestrian's data set in the target image；
Character attribute label is set, collects the pedestrian sample in pedestrian's data set, and institute is added to pedestrian's positive sample
State character attribute label；
Obtain the contextual data collection in the target image；
Scene set attribute tags collect the scene sample that the contextual data is concentrated, and add institute to scene positive sample
State scene properties label；
According to attribute tags pedestrian's positive sample and the scene positive sample generative semantics training set；
By the semantic training set and pedestrian's data set in the form of different task, input described multipair as convolution depth
Degree network model is trained, to obtain the pedestrian in the target image.
In conjunction with the first embodiment of first aspect of the embodiment of the present invention, in second embodiment of the embodiment of the present invention, if
Character attribute label, the pedestrian sample collected in pedestrian's data set are set, and the personage is added to pedestrian's positive sample
Attribute tags, comprising:
The positive negative sample in pedestrian's data set is collected, by the positive and negative sample tissue into two tree structures；
According in the child node of tree structure output, two child nodes with pre-determined distance obtain the pedestrian
Positive sample and pedestrian's negative sample add the character attribute label to pedestrian's positive sample.
In conjunction with the first embodiment and second embodiment of first aspect of the embodiment of the present invention, third of the embodiment of the present invention
It is described multipair as convolution depth network model includes TA-CNN frame in embodiment；
The network structure of the TA-CNN frame, including four convolutional layers, four maximum pond layers and two full articulamentums.
It is described to pass through convolutional Neural in conjunction with the embodiment of the present invention in a first aspect, in the 4th embodiment of the embodiment of the present invention
Certain objects in network VGG19 recognition target image, and certain objects addition second is known in the target image
Other frame, comprising:
The samples pictures of the certain objects are collected, and are marked；
The samples pictures are cleaned and marked；
By the class label and VGG19 of editor, the certain objects in the target image are identified.
Second aspect of the embodiment of the present invention provides a kind of pedestrian detecting system, comprising:
Pedestrian's identification module, for by the multipair pedestrian as in convolution depth network model recognition target image, and
To pedestrian addition the first identification frame in the target image；
Wherein, the multipair training mission as convolution depth network model includes semantic task and pedestrian detection；
Special article identification module, for identifying the specific object in the target image by convolutional neural networks VGG19
Body, and to certain objects addition the second identification frame in the target image；
Monitoring module, if there is overlapping, is sentenced for judging that it is non-overlapping that the first identification frame and the second identification frame have
The fixed pedestrian carries the certain objects, and causes preset monitor event.
In conjunction with second aspect of the embodiment of the present invention, in first embodiment of the embodiment of the present invention, pedestrian's identification module
Include:
Pedestrian's data set acquiring unit, for obtaining pedestrian's data set in the target image；
Pedestrian's positive sample marking unit collects pedestrian's sample in pedestrian's data set for character attribute label to be arranged
This, and the character attribute label is added to pedestrian's positive sample；
Contextual data collection acquiring unit, for obtaining the contextual data collection in the target image；
Scene positive sample marking unit is used for scene set attribute tags, collects the scene sample that the contextual data is concentrated
This, and the scene properties label is added to scene positive sample；
Semantic training set generation unit, for according to pedestrian's positive sample and the positive sample of the scene with attribute tags
This generative semantics training set；
Model training unit, for by the semantic training set and pedestrian's data set in the form of different task, it is defeated
Enter it is described multipair as convolution depth network model is trained, to obtain the pedestrian in the target image.
In conjunction with the first embodiment of second aspect of the embodiment of the present invention, in second embodiment of the embodiment of the present invention, institute
Pedestrian's positive sample marking unit is stated, is also used to:
The positive negative sample in pedestrian's data set is collected, by the positive and negative sample tissue into two tree structures；
According in the child node of tree structure output, two child nodes with pre-determined distance obtain the pedestrian
Positive sample and pedestrian's negative sample add the character attribute label to pedestrian's positive sample.
The third aspect of the embodiment of the present invention provides a kind of terminal device, including memory, processor and is stored in
In above-mentioned memory and the computer program that can be run on above-mentioned processor, when above-mentioned processor executes above-mentioned computer program
The step of realizing method provided by first aspect as above.
The fourth aspect of the embodiment of the present invention provides a kind of computer readable storage medium, above-mentioned computer-readable storage
Media storage has computer program, and above-mentioned computer program realizes method provided by first aspect as above when being executed by processor
The step of.
The embodiment of the present invention proposes a kind of pedestrian detection method, by multipair as convolution depth network model, carries out semantic
The deep learning of task and pedestrian detection, the accurately pedestrian in recognition target image, and in the target image with the first identification
Collimation mark goes out pedestrian；Also by the certain objects in convolutional neural networks VGG19 recognition target image, and in the target image with the
Two identification collimation marks go out certain objects, expand the test object in pedestrian detection, then pass through the first identification frame and the second identification frame
Position, detect and whether there is carrying relationship between pedestrian and special article, cause if pedestrian carries special article default
Monitor event, so that monitoring side is handled in time monitor event.Pedestrian detection method provided in an embodiment of the present invention
So that pedestrian detection is no longer confined in simple character image identification, but by multipair as convolution depth network model and spy
Determine Articles detecting, realizes accurate pedestrian detection, and the monitoring of various pedestrian's attribute, scene properties.
Detailed description of the invention
Fig. 1 is the implementation process schematic diagram for the pedestrian detection method that the embodiment of the present invention one provides；
Fig. 2 is the detailed implementation process schematic diagram of step S102 in Fig. 1；
Fig. 3 is the implementation process schematic diagram of pedestrian detection method provided by Embodiment 2 of the present invention；
Fig. 4 is the schematic diagram of two tree structures provided by Embodiment 2 of the present invention；
Fig. 5 is the character attribute label and scene properties label that the embodiment of the present invention three provides；
Fig. 6 is that the embodiment of the present invention five provides the structural schematic diagram of pedestrian detecting system.
The embodiments will be further described with reference to the accompanying drawings for the realization, the function and the advantages of the object of the present invention.
Specific embodiment
It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not intended to limit the present invention.
It should be noted that, in this document, the terms "include", "comprise" or its any other variant are intended to non-row
His property includes, so that the process, method, article or the device that include a series of elements not only include those elements, and
And further include other elements that are not explicitly listed, or further include for this process, method, article or device institute it is intrinsic
Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including being somebody's turn to do
There is also other identical elements in the process, method of element, article or device.
Herein, using the suffix for indicating such as " module ", " component " or " unit " of element only for advantageous
In explanation of the invention, there is no specific meanings for itself.Therefore, " module " can be used mixedly with " component ".
In subsequent description, inventive embodiments serial number is for illustration only, does not represent the advantages or disadvantages of the embodiments.
Embodiment one
As shown in Figure 1, the embodiment of the invention provides a kind of pedestrian detection methods, first by multipair as convolution depth net
Network model improves the accuracy of pedestrian detection, then in conjunction with convolutional neural networks VGG19, realizes the detection function of special article
Energy.Method includes but is not limited to following steps:
S101, by the multipair pedestrian as in convolution depth network model recognition target image, and in the target image
In to the pedestrian addition first identification frame.
It is multipair as convolution depth network model in above-mentioned steps S101, multiple tasks and multiple numbers are come from for learning
Attribute label is carried out according to the advanced features in source, and to validity feature therein, thus by attribute information from existing scene cut
For data set, it is transferred to pedestrian's data set.
In embodiments of the present invention, multipair as carrying out in convolution depth network model for task includes semantic task and pedestrian
Detection obtains semantic information by semantic task training, and semantic information is usually the attribute for the target being trained to, while passing through language
Adopted information assists carry out pedestrian detection, to reduce miss rate of the deep learning method in pedestrian detection.
S102, the certain objects in the target image are identified by convolutional neural networks VGG19, and in the target figure
To certain objects addition the second identification frame as in.
In above-mentioned steps S102, convolutional neural networks VGG19, for carrying out picture recognition and classification.
In embodiments of the present invention, whether there are certain objects in recognition target image, needing first to advance to certain objects identifies,
And store recognition result.
In one embodiment, as shown in Fig. 2, above-mentioned steps S102 may include:
S1021, the samples pictures for collecting the certain objects, and be marked.
In above-mentioned steps S1021, the picture that image data may come from kaggle match, *** or Baidu is searched
Rope, ImageNet image library etc..
S1022, the samples pictures are cleaned and is marked.
In above-mentioned steps S1022, cleaning specifically can be, but not limited to include: to remove undesirable picture, interception
The part needed, the size and clarity of interception need to unitize, and carry out class label to the image after cleaning, such as carry on the back
Doubtful dangerous objects such as packet, cap, club etc..
S1023, class label and VGG19 by editor, identify the certain objects in the target image.
In above-mentioned steps S1023, special article can be the article of dangerous goods or doubtful dangerous goods；It is also possible to
Such as article of laptop, luggage valuables or doubtful valuables.
It,, then can be by belongings to certain objects addition the second identification frame by identifying special article in important events
Product go to judge the behavior of pedestrian, or judge scenario by special article, so that the content of abundant monitoring, improves pedestrian detection
Practical significance.
For example, when recognizing perambulator, it can be determined that this pedestrian has child；When recognizing fire, it can be determined that work as front court
There may be fire for scape.
S103, judge that the first identification frame and described second identifies that frame has non-overlapping, if there is overlapping, determine the row
People carries the certain objects, and causes preset monitor event.
In above-mentioned steps S103, preset monitor event can be, but not limited to include: transmission pre-alert notification, control camera shooting
Head carries out continuing to track and record a video to the people.
Pedestrian detection method provided in an embodiment of the present invention carries out semantic appoint by multipair as convolution depth network model
The deep learning of business and pedestrian detection, the accurately pedestrian in recognition target image, and frame is identified with first in the target image
Mark pedestrian；Also by the certain objects in convolutional neural networks VGG19 recognition target image, and in the target image with second
Identification collimation mark goes out certain objects, expands the test object in pedestrian detection, then identifies frame by the first identification frame and second
Position is detected and whether there is carrying relationship between pedestrian and special article, causes if pedestrian carries special article preset
Monitor event enables monitoring side to handle in time monitor event.Pedestrian detection method provided in an embodiment of the present invention makes
Pedestrian detection is obtained no longer to be confined in simple character image identification, but by multipair as convolution depth network model and specific
Articles detecting realizes accurate pedestrian detection, and the monitoring of various pedestrian's attribute, scene properties.
Embodiment two
As shown in figure 3, the embodiment of the present invention is shown in embodiment one about the multipair as convolution depth net of step S101
The identification process of network model.In embodiments of the present invention, step S101 includes but is not limited to:
Pedestrian's data set in S1011, the acquisition target image.
In above-mentioned steps S1011, applying in multipair pedestrian's data set as convolution depth network model is image data
Collection, image data set is that the subgraph photo based on image is obtained.
In embodiments of the present invention, before obtaining pedestrian's data set, block processing first is carried out to target image, then after processing
Subgraph photo in obtain data about personage.It only include the subgraph with character data then in acquired pedestrian's data set
Photo.
In a particular application, how much related the size of each subgraph photo and each character data be.
S1012, setting character attribute label, collect the pedestrian sample in pedestrian's data set, and to pedestrian's positive sample
Add the character attribute label.
In above-mentioned steps S1012, pedestrian sample be expert at personal data concentrate extract feature samples, with character attribute
The relevant feature samples of label are then pedestrian's positive sample.
Character attribute label can include but is not limited to, knapsack, cap, white clothes etc..
Contextual data collection in S1013, the acquisition target image.
S1014, scene set attribute tags collect the scene sample that the contextual data is concentrated, and to scene positive sample
Add the scene properties label.
Similarly, it only includes having contextual data that acquired contextual data, which is concentrated, by above-mentioned steps S1013 and step S1014
Subgraph photo, feature samples relevant to scene properties label are then scene positive sample.
Scene properties label can include but is not limited to, sky, tree, building etc..
S1015, basis pedestrian's positive sample and the scene positive sample generative semantics training set with attribute tags.
In above-mentioned steps S1015, semantic training set shows pedestrian's attribute of target image by pedestrian's positive sample, passes through
The scene properties of scene positive sample performance data set.
S1016, by the semantic training set and pedestrian's data set in the form of different task, input it is described it is multipair as
Convolution depth network model is trained, to obtain the pedestrian in the target image.
In a particular application, when general depth network model carries out pedestrian detection, pedestrian detection task is considered as single
Binary classification task, positive sample may be made to obscure with a large amount of negative samples.
In one embodiment, above-mentioned steps S1012 may include:
Character attribute label is set, collects the positive negative sample in pedestrian's data set, the positive and negative sample tissue is arrived
In two tree structures；
According to two child nodes in the child node of tree structure output with pre-determined distance, the pedestrian is being obtained just
Sample and pedestrian's negative sample add the character attribute label to pedestrian's positive sample.
As shown in figure 4, the embodiment of the present invention also proposed the schematic diagram of two tree structures, in two tree structures:
Each father node: the HOG (Histogram of Oriented Gradient, histograms of oriented gradients) for extracting positive negative sample is special
Sign, and assemble data using K mean cluster algorithm；Each child node: clustering father node, the vector space between node
Structure is obtained by the mean value of series connection distance and each leaf node.
By two tree structures, pedestrian detection is divided into two classification tasks according to positive negative sample, wherein positive sample
Feature and the degree of correlation of character attribute label are high, and the feature of negative sample is low with the degree of correlation of character attribute label, thus preliminarily
Positive negative sample is separated, then further according to the distance between two child nodes, pedestrian's positive sample is selected, avoids pedestrian in positive sample
When the feature of positive sample is low with the degree of correlation of character attribute label, and the feature of pedestrian's negative sample and character attribute mark in negative sample
The high situation of the degree of correlation of label, so that the problem of pedestrian's positive sample and pedestrian's negative sample are obscured.
Embodiment three
The embodiment of the present invention is with Caltech (P) data set and CamVid (Ba), Stanford Background (Bb), LM
For+SUN (Bc) data set, illustrate that semanteme training set shown in step S1011 to step S1015 obtained in embodiment two
Journey.
Pedestrian sample is collected from Caltech (P) first, pedestrian's positive sample is marked using 9 class character attribute labels, people
Object attribute tags are mainly provided by the UK police for being monitored analysis.
Then, from CamVid (Ba), Stanford Background (Bb) collects environment on LM+SUN (Bc) data set
Scene sample, scene positive sample are marked using 8 class scene properties labels.As shown in figure 5, the embodiment of the present invention also shows reality
In the application of border, it is usually required mainly for the character attribute label and scene properties label considered.
Then, the form by label of generative semantics training set exports:
Wherein,Classification from as shown in Figure 5
Label classification, p indicate that character attribute label, s and u indicate that environment attribute label, n indicate the position letter of data set neutron image piece
Breath, N indicate the clique photo quantity in data set.
Above-mentioned semantic training set includes all subgraph photos of target image, and pedestrian detection and semantic may be implemented
The frame that the multitask of business learns jointly.
Example IV
The embodiment of the present invention in above-described embodiment one and embodiment two it is multipair as convolution depth network architecture into
Row explanation.
In embodiments of the present invention, multipair as convolution depth network model includes TA-CNN (task-assistant
Convolutional Neural Networks, task assistant convolutional neural networks) frame；The network of the TA-CNN frame
Structure, including four convolutional layers, four maximum pond layers and two full articulamentums.
In embodiments of the present invention, TA-CNN frame is the Alex Net of simplified version, is removed on the basis of former Alex Net
One layer of convolutional layer and full articulamentum, and joined SPV (Structure Projection Vector, vector space structure).
The then TA-CNN frame in the embodiment of the present invention, four included convolutional layers are conv1 to conv4, two full articulamentums
For fc5 and fc6.
As it can be seen that the multipair building basis as convolution depth network model in above-described embodiment one and embodiment two is TA-
CNN frame also needs to optimize TA-CNN frame in a particular application, and the embodiment of the present invention is also in embodiment three as a result,
Semantic data collection for, show the optimization process of TA-CNN frame:
Firstly, formulation TA-CNN, TA-CNN are the following log posterior probability of optimization:
In order to solve Caltech (P), CamVid (Ba), Stanford Background (Bb), LM+SUN (Bc) each number
According to the gap between collection, each sample x is calculatednStructuring projection vector zn, loss function becomes:
Then pass through TA-CNN e-learning, wherein for learning network parameter W, (2) formula is formulated as again
Softmax loss function, it may be assumed that
As it can be seen that formula (3) puts 8 loss functions optimization together, but it will lead to two problems:
1) different task rates of convergence is different, and training will lead to over-fitting simultaneously for multitask；
If 2) characteristic dimension is relatively high, the parameter of network high level can be very much.
In the embodiment of the present invention, in order to solve the problems, such as above-mentioned two, multivariate is converted by (3) formula
Cross-entropy loss, formula are as follows:
Embodiment five
As shown in fig. 6, the embodiment of the invention provides a kind of pedestrian detecting systems 60, comprising:
Pedestrian's identification module 61, for by the multipair pedestrian as in convolution depth network model recognition target image, and
In the target image to pedestrian's addition the first identification frame；
Wherein, the multipair training mission as convolution depth network model includes semantic task and pedestrian detection；
Special article identification module 62, for passing through the certain objects in convolutional neural networks VGG19 recognition target image,
And in the target image to certain objects addition the second identification frame；
Monitoring module 63, if there is overlapping, determines pedestrian for judging that it is non-overlapping that the first identification frame and the second identification frame have
Certain objects are carried, and cause preset monitor event.
In one embodiment, pedestrian's identification module 61 includes:
Pedestrian's data set acquiring unit, for obtaining pedestrian's data set in target image；
Pedestrian's positive sample marking unit collects the pedestrian sample in pedestrian's data set for being arranged character attribute label, and
Character attribute label is added to pedestrian's positive sample；
Contextual data collection acquiring unit, for obtaining the contextual data collection in target image；
Scene positive sample marking unit is used for scene set attribute tags, collects the scene sample in scene data set, and
Scene properties label is added to scene positive sample；
Semantic training set generation unit, for generating language according to pedestrian's positive sample and scene positive sample with attribute tags
Adopted training set；
Model training unit, for by semantic training set and pedestrian's data set in the form of different task, input it is multipair as
Convolution depth network model is trained, to obtain the pedestrian in target image.
In embodiments of the present invention, pedestrian's positive sample marking unit, is also used to:
The positive negative sample in pedestrian's data set is collected, by positive and negative sample tissue into two tree structures；
According to tree structure output child node in, with pre-determined distance two child nodes, obtain pedestrian's positive sample and
Pedestrian's negative sample adds character attribute label to pedestrian's positive sample.
The embodiment of the present invention also provide a kind of terminal device include memory, processor and storage on a memory and can be
The computer program run on processor when the processor executes the computer program, is realized as described in embodiment one
Pedestrian detection method in each step.
The embodiment of the present invention also provides a kind of storage medium, and the storage medium is computer readable storage medium, thereon
It is stored with computer program, when the computer program is executed by processor, realizes the pedestrian detection as described in embodiment one
Each step in method.
Embodiment described above is merely illustrative of the technical solution of the present invention, rather than its limitations；Although previous embodiment
Invention is explained in detail, those skilled in the art should understand that: it still can be to aforementioned each implementation
Technical solution documented by example is modified or equivalent replacement of some of the technical features；And these modification or
Replacement, the spirit and scope for technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution should all include
Within protection scope of the present invention.
Claims (10)
1. a kind of pedestrian detection method characterized by comprising
By the multipair pedestrian as in convolution depth network model recognition target image, and to the row in the target image
People's addition the first identification frame；
Wherein, the multipair training mission as convolution depth network model includes semantic task and pedestrian detection；
The certain objects in the target image are identified by convolutional neural networks VGG19, and to institute in the target image
State certain objects addition the second identification frame；
It is non-overlapping to judge that the first identification frame and the second identification frame have, if there is overlapping, determines that the pedestrian carries
The certain objects, and cause preset monitor event.
2. pedestrian detection method as described in claim 1, which is characterized in that it is described by multipair as convolution depth network model
Pedestrian in recognition target image, comprising:
Obtain pedestrian's data set in the target image；
Character attribute label is set, collects the pedestrian sample in pedestrian's data set, and the people is added to pedestrian's positive sample
Object attribute tags；
Obtain the contextual data collection in the target image；
Scene set attribute tags collect the scene sample that the contextual data is concentrated, and add the field to scene positive sample
Scape attribute tags；
According to attribute tags pedestrian's positive sample and the scene positive sample generative semantics training set；
By the semantic training set and pedestrian's data set in the form of different task, input described multipair as convolution depth net
Network model is trained, to obtain the pedestrian in the target image.
3. pedestrian detection method as claimed in claim 2, which is characterized in that setting character attribute label, described in the collection
Pedestrian sample in pedestrian's data set, and the character attribute label is added to pedestrian's positive sample, comprising:
The positive negative sample in pedestrian's data set is collected, by the positive and negative sample tissue into two tree structures；
According in the child node of tree structure output, two child nodes with pre-determined distance obtain the positive sample of the pedestrian
Originally with pedestrian's negative sample, the character attribute label is added to pedestrian's positive sample.
4. pedestrian detection method as described in any one of claims 1 to 3, which is characterized in that described multipair as convolution depth net
Network model includes task assistant's convolutional neural networks TA-CNN frame；
The network structure of the TA-CNN frame, including four convolutional layers, four maximum pond layers and two full articulamentums.
5. pedestrian detection method as described in claim 1, which is characterized in that described to be identified by convolutional neural networks VGG19
Certain objects in target image, and to certain objects addition the second identification frame in the target image, comprising:
The samples pictures of the certain objects are collected, and are marked；
The samples pictures are cleaned and marked；
By the class label and VGG19 of editor, the certain objects in the target image are identified.
6. a kind of pedestrian detecting system characterized by comprising
Pedestrian's identification module, for by the multipair pedestrian as in convolution depth network model recognition target image, and described
To pedestrian addition the first identification frame in target image；
Wherein, the multipair training mission as convolution depth network model includes semantic task and pedestrian detection；
Special article identification module, for identifying the certain objects in the target image by convolutional neural networks VGG19, and
To certain objects addition the second identification frame in the target image；
Monitoring module, if there is overlapping, determines institute for judging that it is non-overlapping that the first identification frame and the second identification frame have
It states pedestrian and carries the certain objects, and cause preset monitor event.
7. pedestrian detecting system as claimed in claim 6, which is characterized in that pedestrian's identification module includes:
Pedestrian's data set acquiring unit, for obtaining pedestrian's data set in the target image；
Pedestrian's positive sample marking unit collects the pedestrian sample in pedestrian's data set for character attribute label to be arranged, and
The character attribute label is added to pedestrian's positive sample；
Contextual data collection acquiring unit, for obtaining the contextual data collection in the target image；
Scene positive sample marking unit is used for scene set attribute tags, collects the scene sample that the contextual data is concentrated, and
The scene properties label is added to scene positive sample；
Semantic training set generation unit, for raw according to pedestrian's positive sample and the scene positive sample with attribute tags
At semantic training set；
Model training unit, for the semantic training set and pedestrian's data set in the form of different task, to be inputted institute
State it is multipair as convolution depth network model is trained, to obtain the pedestrian in the target image.
8. pedestrian detecting system as claimed in claim 7, which is characterized in that pedestrian's positive sample marking unit is also used to:
The positive negative sample in pedestrian's data set is collected, by the positive and negative sample tissue into two tree structures；
According in the child node of tree structure output, two child nodes with pre-determined distance obtain the positive sample of the pedestrian
Originally with pedestrian's negative sample, the character attribute label is added to pedestrian's positive sample.
9. a kind of terminal device, which is characterized in that on a memory and can be on a processor including memory, processor and storage
The computer program of operation, which is characterized in that when the processor executes the computer program, realize such as claim 1 to 5
Each step in described in any item pedestrian detection methods.
10. a kind of storage medium, the storage medium is computer readable storage medium, is stored thereon with computer program,
It is characterized in that, when the computer program is executed by processor, realizes such as pedestrian detection described in any one of claim 1 to 5
Each step in method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910397924.2A CN110245564B (en) | 2019-05-14 | 2019-05-14 | Pedestrian detection method, system and terminal equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910397924.2A CN110245564B (en) | 2019-05-14 | 2019-05-14 | Pedestrian detection method, system and terminal equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110245564A true CN110245564A (en) | 2019-09-17 |
| CN110245564B CN110245564B (en) | 2024-07-09 |
Family
ID=67884440
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910397924.2A Active CN110245564B (en) | 2019-05-14 | 2019-05-14 | Pedestrian detection method, system and terminal equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110245564B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111178403A (en) * | 2019-12-16 | 2020-05-19 | 北京迈格威科技有限公司 | Method and device for training attribute recognition model, electronic equipment and storage medium |
| CN111553228A (en) * | 2020-04-21 | 2020-08-18 | 佳都新太科技股份有限公司 | Method, device, equipment and storage medium for detecting personal bag relationship |
| CN111881791A (en) * | 2020-07-16 | 2020-11-03 | 北京宙心科技有限公司 | Security identification method and system |
| CN112418043A (en) * | 2020-11-16 | 2021-02-26 | 安徽农业大学 | Corn weed occlusion determination method and device, robot, equipment and storage medium |
| CN112837454A (en) * | 2021-01-28 | 2021-05-25 | 深圳市商汤科技有限公司 | Passage detection method and device, electronic equipment and storage medium |
| WO2022227772A1 (en) * | 2021-04-27 | 2022-11-03 | 北京百度网讯科技有限公司 | Method and apparatus for training human body attribute detection model, and electronic device and medium |
| CN115527158A (en) * | 2022-08-11 | 2022-12-27 | 北京市燃气集团有限责任公司 | Method and device for detecting abnormal behaviors of personnel based on video monitoring |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160180195A1 (en) * | 2013-09-06 | 2016-06-23 | Toyota Jidosha Kabushiki Kaisha | Augmenting Layer-Based Object Detection With Deep Convolutional Neural Networks |
| CN106485268A (en) * | 2016-09-27 | 2017-03-08 | 东软集团股份有限公司 | A kind of image-recognizing method and device |
| CN108875501A (en) * | 2017-11-06 | 2018-11-23 | 北京旷视科技有限公司 | Human body attribute recognition approach, device, system and storage medium |
| CN109359515A (en) * | 2018-08-30 | 2019-02-19 | 东软集团股份有限公司 | A kind of method and device that the attributive character for target object is identified |
-
2019
- 2019-05-14 CN CN201910397924.2A patent/CN110245564B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160180195A1 (en) * | 2013-09-06 | 2016-06-23 | Toyota Jidosha Kabushiki Kaisha | Augmenting Layer-Based Object Detection With Deep Convolutional Neural Networks |
| CN106485268A (en) * | 2016-09-27 | 2017-03-08 | 东软集团股份有限公司 | A kind of image-recognizing method and device |
| CN108875501A (en) * | 2017-11-06 | 2018-11-23 | 北京旷视科技有限公司 | Human body attribute recognition approach, device, system and storage medium |
| CN109359515A (en) * | 2018-08-30 | 2019-02-19 | 东软集团股份有限公司 | A kind of method and device that the attributive character for target object is identified |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111178403A (en) * | 2019-12-16 | 2020-05-19 | 北京迈格威科技有限公司 | Method and device for training attribute recognition model, electronic equipment and storage medium |
| CN111178403B (en) * | 2019-12-16 | 2023-10-17 | 北京迈格威科技有限公司 | Method, device, electronic equipment and storage medium for training attribute identification model |
| CN111553228A (en) * | 2020-04-21 | 2020-08-18 | 佳都新太科技股份有限公司 | Method, device, equipment and storage medium for detecting personal bag relationship |
| CN111553228B (en) * | 2020-04-21 | 2021-10-01 | 佳都科技集团股份有限公司 | Method, device, equipment and storage medium for detecting personal bag relationship |
| CN111881791A (en) * | 2020-07-16 | 2020-11-03 | 北京宙心科技有限公司 | Security identification method and system |
| CN111881791B (en) * | 2020-07-16 | 2021-10-15 | 北京宙心科技有限公司 | Security identification method and system |
| CN112418043A (en) * | 2020-11-16 | 2021-02-26 | 安徽农业大学 | Corn weed occlusion determination method and device, robot, equipment and storage medium |
| CN112418043B (en) * | 2020-11-16 | 2022-10-28 | 安徽农业大学 | Corn weed occlusion determination method and device, robot, equipment and storage medium |
| CN112837454A (en) * | 2021-01-28 | 2021-05-25 | 深圳市商汤科技有限公司 | Passage detection method and device, electronic equipment and storage medium |
| WO2022227772A1 (en) * | 2021-04-27 | 2022-11-03 | 北京百度网讯科技有限公司 | Method and apparatus for training human body attribute detection model, and electronic device and medium |
| CN115527158A (en) * | 2022-08-11 | 2022-12-27 | 北京市燃气集团有限责任公司 | Method and device for detecting abnormal behaviors of personnel based on video monitoring |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110245564B (en) | 2024-07-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110245564A (en) | A kind of pedestrian detection method, system and terminal device | |
| Tian et al. | Multimodal deep representation learning for video classification | |
| Xu et al. | Video structured description technology based intelligence analysis of surveillance videos for public security applications | |
| Yuen et al. | A data-driven approach for event prediction | |
| Malgireddy et al. | Language-motivated approaches to action recognition | |
| US8842965B1 (en) | Large scale video event classification | |
| EP3765995B1 (en) | Systems and methods for inter-camera recognition of individuals and their properties | |
| Hu et al. | Video structural description technology for the new generation video surveillance systems | |
| Kaliyar et al. | A Hybrid Model for Effective Fake News Detection with a Novel COVID-19 Dataset. | |
| CN110972499A (en) | Labeling system of neural network | |
| Ding et al. | Prior knowledge-based deep learning method for indoor object recognition and application | |
| Tian et al. | MCA-NN: Multiple correspondence analysis based neural network for disaster information detection | |
| Alonso-Bartolome et al. | Multimodal fake news detection | |
| CN112200176A (en) | Method and system for detecting quality of face image and computer equipment | |
| CN116975340A (en) | Information retrieval method, apparatus, device, program product, and storage medium | |
| CN111027622A (en) | Picture label generation method and device, computer equipment and storage medium | |
| Gautam et al. | Discrimination and detection of face and non-face using multilayer feedforward perceptron | |
| Al-Jamal et al. | Image captioning techniques: a review | |
| CN112507912B (en) | Method and device for identifying illegal pictures | |
| Abdallah et al. | Multilevel deep learning-based processing for lifelog image retrieval enhancement | |
| Wang et al. | A lightweight CNN model based on GhostNet | |
| Liu et al. | Determining the best attributes for surveillance video keywords generation | |
| Tao et al. | Florida international university-university of miami trecvid 2019 | |
| Shi et al. | Uncertain and biased facial expression recognition based on depthwise separable convolutional neural network with embedded attention mechanism | |
| CN113297934A (en) | Multi-mode video behavior analysis method for detecting internet violent harmful scene |
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 |