CN105160649A - Multi-target tracking method and system based on kernel function unsupervised clustering - Google Patents
Multi-target tracking method and system based on kernel function unsupervised clustering Download PDFInfo
- Publication number
- CN105160649A CN105160649A CN201510386915.5A CN201510386915A CN105160649A CN 105160649 A CN105160649 A CN 105160649A CN 201510386915 A CN201510386915 A CN 201510386915A CN 105160649 A CN105160649 A CN 105160649A
- Authority
- CN
- China
- Prior art keywords
- target
- point
- module
- ground
- camera
- 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.)
- Pending
Links
Classifications
-
- 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
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to the image processing field and relates to a multi-target tracking method and a system based on kernel function unsupervised clustering. According to the method, a binocular camera is utilized to acquire left and right sequence images at one same time, and parameters of the binocular camera are utilized for image correction; a parallax error is calculated through extracting image characteristic points and matching characteristics; the acquired parallax error is utilized to calculate the coordinate position of a target characteristic point relative to the camera, namely the coordinate of the camera, ground calibration is accomplished, ground shadow characteristic points can be filtered according to height from the characteristic point to the ground, and ground shadow interference is eliminated; according to the three-dimensional coordinate characteristic point, in combination with the kernel function, unsupervised clustering is carried out for targets with undetermined category quantity, all characteristic points of one target are gathered to form one set, one category corresponds to the position and the direction of one observation value, a present frame of the target can be acquired in combination with the position and the direction of the previous frame target, namely the prediction position value and the prediction direction value, an optimum estimation algorithm is utilized to acquire the position and the direction of the optimum target, and thereby the multi-target fast tracking effect is realized.
Description
Technical field
What the present invention relates to is a kind of technical field of image processing Multitarget Tracking and system, specifically a kind of Multitarget Tracking based on kernel function Non-surveillance clustering, and realizes the software systems of this technology.
Background technology
Along with the development of computer vision and people constantly strengthen public security consciousness, multi-object monitoring has occupied more and more consequence in production and life.Object Detecting and Tracking has become important research contents.In supervisory system, track algorithm can reduce cost of labor, saves social resources.But because have a lot of probabilistic factor in tracking environmental, the problems such as the diversity of target, photoenvironment complicated and changeable, shadow interference all can bring interference to track algorithm.This also causes the work that current existing a lot of system cannot be stable under actual application environment, thus lacks intelligent monitor system truly.For multiple target tracking, except the problems referred to above, the correlativity between each target, mutually the problem such as to block and all can bring larger difficulty to accurately following the tracks of.
Based on this, Shandong University teacher Chang Faliang and teacher Ke Jing based on the moving object detection and tracking of binocular vision can effectively solve block, the problem such as shade.2006, teacher Chang Faliang was by the Camshift algorithm application in monocular in binocular vision, and proposition in 2007 is followed the tracks of moving target based on the solid matching method of hierarchical network smallest partition.Teacher Ke Jing is from the Harris angle point algorithms to improve Detection results improved afterwards, then, they utilize again before the accurate angle point of Harris coupling as reference mark, the coupling of carrying out being correlated with in region obtains overall dense disparity map.But dense disparity map computation complexity is still very high like this, especially when scene background complexity is higher, be difficult to the effect reaching real-time follow-up.
Through finding the retrieval of prior art, open (bulletin) the day 2013.05.15 of Chinese patent literature CN103106659A, disclose a kind of depletion region object detecting and tracking method based on the sparse Point matching of binocular vision, also be utilize biocular systems to obtain unique point world coordinate system, and be projected on ground level, complete discrete point cluster by buman body type information and golden section.Finally, take cluster as unit, complete the robust detection and tracking of guarded region one skilled in the art target in conjunction with JPDA and target color information.This technology makes operand diminish, and makes target detection more accurate with tracking.But this technology is by after projecting characteristic points in world coordinate system to ground, utilize buman body type and golden section comparison two-dimensional projection's point to carry out cluster, be difficult to accomplish that the high precision to objective shape is mated.In addition, use joint probability data to the first frame target sparse unique point convex polygon color histogram and target update target location newly detected, the calculating of dense Region has in fact been got back to again in histogrammic statistical computation, just reduce the computer capacity of dense Region, when target is more, be still difficult to requirement of real time.
Open (bulletin) the day 2009.01.14 of Chinese patent literature CN101344965, disclose a kind of full automatic object detecting and tracking system of computer vision field, wherein: load module is responsible for gathering digital picture captured by binocular camera and is inputted as system, the digital picture obtained is input to characteristic extracting module and carries out signature analysis to wherein a kind of image and go out some unique points as process image subsequently.Calculating its parallax by mating in two images after unique point, in conjunction with the camera inside and outside parameter known in advance, coordinate under the camera coordinates system of unique point can be calculated.Further by the relation of world coordinate system and camera coordinates system, coordinate under known unique point world coordinates.These feature points clusterings become set to express target location by cluster module, and the target location in trajectory analysis module sequence estimated time draws the movement locus of target.But this technology cluster utilizes target signature point height and position to carry out cluster, faces aimed at precision matching problem equally, when background more complicated is, be difficult to mate accurately difform target.
Summary of the invention
The present invention is directed to prior art above shortcomings, propose a kind of multi-object tracking method based on kernel function Non-surveillance clustering, and realize this technical software system.Utilize multi-cam to obtain image, by the contact between different view field image target signature, calculate unique point volume coordinate, solve the interference that occlusion issue brings.Then, by extracting target sparse unique point and calculating the three-dimensional information in locus, and use invent herein based on kernel function Non-surveillance clustering method, using target location and target direction as variable, directly world coordinates three-dimensional feature point is climbed the mountain and cluster, determine multiobject position and direction.And in conjunction with optimal estimation algorithm, according to previous frame tracking results and present frame testing result, assessment optimal objective position and direction.The kernel function that the present invention introduces according to different characteristic, kernel function that difform target formation is different, can be mated different target more accurately.In addition, by optimal estimation algorithm, the target location only using sparse features point cluster result to obtain and the result of direction and previous frame just can estimate present frame optimal objective position and direction.
The present invention is achieved by the following technical solutions:
The present invention relates to a kind of multi-object tracking method based on kernel function Non-surveillance clustering, comprise the following steps:
1) first obtain the left and right sequence chart of synchronization as input by binocular camera, utilize binocular camera parameter to image rectification.
2) by extracting image characteristic point and matching characteristic, parallax is calculated.
3) utilize the relative camera coordinates position of disparity computation target signature point obtained, i.e. camera coordinates, and complete ground demarcation, thus according to unique point distance floor level, ground shadow character point can be filtered, eliminate the interference of ground area shading.
4) for three-dimensional coordinate unique point, be combined kernel function, Non-surveillance clustering is carried out to the target of uncertain classification number, all unique points of a target are aggregated into a set, the i.e. position of the corresponding observed reading of each classification and direction, and obtain target place present frame in conjunction with the position of previous frame target and direction prediction, the i.e. position of predicted value and direction prediction value, use optimal estimation algorithm finally to obtain position and the direction of optimal objective, thus reach the effect that multiple goal follows the tracks of fast.
Described image rectification refers to the inside and outside parameter utilizing binocular camera, to the correct image got, makes:
I) one, synchronization space point meets Epipolar geometry theorem in left and right sequence chart.
Ii) determine that a corresponding locus is on the polar curve of another sequence chart for one in the arbitrary width in left and right sequence chart.
Described matching characteristic refers to extract minutiae in the arbitrary width first in left and right sequence chart, and utilizes the impact point polar curve of matching algorithm in another sequence chart being searched for correspondence.
Described ground is demarcated and is referred to: make marks on the ground, extract the unique point on this mark in left and right sequence chart, calculate its camera coordinates, and utilize least square method to ground unique point camera coordinates matching ground level parametric equation.
Described step 4, specifically comprises:
A) all unique points are filtered, filter out and highly may belong to clarification of objective point, then be each unique point initialization angle θ
0, the anglec of rotation of certain target may be belonged to as each unique point, i.e. target direction;
B) construct the weighted value that kernel density function is each sample point estimated probability, find target location in the corresponding space of local maximum at probability space.When searching for, the probability space point that constantly search is larger in its contiguous range from a starting point, until reach the position of local maximum point.Namely mean ?shift hill climbing process, and use Non-surveillance clustering method to carry out cluster to the characteristic point position after climbing the mountain, classification number represents target number.
C) utilize optimal estimation algorithm, the motion vector new according to the motion vector prediction in target past, upper frame position basis obtains the predicted position of present frame, reaches the effect to the correction of current location measured value.
Described structure kernel density function refers to, by characteristic point position variable and angle variables, constructs the probability density of this unique point of function representation at floor projection:
Wherein: the local maximum point in probability space E (x, θ) correspond to the densest position of discrete space mid point, the target that present frame detects, H is represented
θ(d
i(θ)), H
x(d
j(x), θ
i) be the kernel function of direction and position, d
i(θ), d
jx () is the normalized cumulant measure function of direction and position, d
i(θ)=|| A
θ(θ-θ
i) ||
2, d
j(θ)=|| A
x(x-x
j) ||
2, the variable x in measure function
jrefer to jth unique point projection coordinate on the ground, variable θ
iat x
jrepresent i-th direction value in the coordinate [0,2 π] of a jth Projection Character point, w
jrepresent the weighted value of a jth Projection Character point, relative importance when namely Projection Character point determines target location.Also larger near target's center's importance.
Described kernel function H
xwith direction kernel function H
θrepresent the spatial property of target signature point, different target types has different space attributes, kernel function that also should be corresponding different, H
x(d
j(x), θ
i) refer at direction parameter θ
ieffect move down elliptical center by H
x(d
j) be rotated counterclockwise θ
ithe kernel function that angle obtains, direction kernel function represents the anglec of rotation of shape kernel function relative centre, in [0,2 π] region, evenly choose n
iindividual angle samples value.
Described use Non-surveillance clustering method is carried out cluster to the characteristic point position after climbing the mountain and is referred to, being gathered by point very little for distance after climbing the mountain is a class, represents a target, wherein: the distance between two unique points can be expressed as d (x
i, x
j), be then divided into a class when distance is less than certain value.
For cluster result, if two little classification distances are very near, then two little classifications are merged into a large classification.Otherwise that divides as unique point in fruit for a large classification relatively opens, then this large classification is split as two classifications.Whether belong to large cluster to evaluate and test cluster and be divided into two little clusters, the variance based on direction distance definition intra-cluster point distance is:
this variance represents less cluster, have comparatively authority re-projection point all near central point, namely the variance of this cluster is less, and large cluster comprises the subpoint of multiple target, cause the point of many large weights all to cause statistical variance larger away from cluster center point, therefore two new clusters should be split into when certain cluster variance is larger.
The present invention relates to a kind of system realizing said method, comprise: camera calibration module, parameter input module, image processing module, feature extracting and matching module, ground demarcating module, ordinate transform module, kernel function cluster module, filter tracking module, showing interface module, wherein: camera calibration module is demarcated the camera parameter of binocular camera and is stored in parameter input module, parameter input module reads the camera parameter demarcated and exports image processing module to, image processing module to correct binocular sequence of pictures according to camera parameter and exports feature extracting and matching module to, feature extracting and matching module extracts target signature from the picture after correction, also ordinate transform module and ground demarcating module is exported respectively to by relevant matches unique point calculating unique point parallax and camera coordinates, ground demarcating module is demarcated ground parameter, and export ground parameter to ordinate transform module when following the tracks of, the camera coordinates of unique point is converted to world coordinates and exports kernel function cluster module to by ordinate transform module carries out cluster, each target is polymerized to a set by kernel function cluster module, namely present frame testing result exports filter tracking module to, filter tracking module estimates optimum target location and direction according to previous frame result and present frame testing result, and show tracking results by showing interface module.
Technique effect
Compared with prior art, the present invention utilizes the binocular vision platform built that target signature in monitoring scene is restored three-dimensional coordinate information.Novel kernel function clustering algorithm forms cluster set according to unique point distance floor level and position, determines the quantity of target, position and direction.The method and tradition based on color characteristic method compared with, more stable under Varying Illumination because two camera looks into fee Same Scene illumination variation can be cancelled out each other.Simultaneously closely the unique point of the shade of board plane, due to highly lower and be filtered, thus efficiently solves the interference that shadow problem brings, and target does not exist the occlusion issue between target under depression angle.In addition, invent herein based on kernel function to sparse features point Non-surveillance clustering, can not only realize multiobject tracking, and utilize sparse features point clustering method filtering algorithm for estimating to avoid the complicated calculations of classic method dense characteristic, make system meet requirement of real-time.
Accompanying drawing explanation
Fig. 1 is system module schematic diagram.
Fig. 2 is schematic flow sheet of the present invention.
Fig. 3 is embodiment schematic diagram 1.
Fig. 4 is embodiment schematic diagram 2.
Embodiment
Elaborate to embodiments of the invention below, the present embodiment is implemented under premised on technical solution of the present invention, give detailed embodiment and concrete operating process, but protection scope of the present invention is not limited to following embodiment.
Embodiment 1
As shown in Figure 1, the system that the present embodiment adopts comprises: camera calibration module, parameter input module, image processing module, feature extracting and matching module, ground demarcating module, ordinate transform module, kernel function cluster module, filter tracking module, showing interface module, wherein: camera calibration module is demarcated the camera parameter of binocular camera and is stored in parameter input module, parameter input module reads the camera parameter demarcated and exports image processing module to, image processing module to correct binocular sequence of pictures according to camera parameter and exports feature extracting and matching module to, feature extracting and matching module extracts target signature from the picture after correction, also ordinate transform module and ground demarcating module is exported respectively to by relevant matches unique point calculating unique point parallax and camera coordinates, ground demarcating module is demarcated ground parameter, and export ground parameter to ordinate transform module when following the tracks of, the camera coordinates of unique point is converted to world coordinates and exports kernel function cluster module to by ordinate transform module carries out cluster, each target is polymerized to a set by kernel function cluster module, namely present frame testing result exports filter tracking module to, filter tracking module estimates optimum target location and direction according to previous frame result and present frame testing result, and show tracking results by showing interface module.
As shown in Figure 2, the present embodiment comprises the following steps:
1) set up binocular equipment, obtained binocular video stream and input to computing machine.
2) by joining inside and outside camera calibration method calibration for cameras, proofread positive binocular image inside and outside utilization, make the image coordinate of unified goal point in two figure more in the same horizontal line, i.e. polar curve level, and carry out down-sampled, for subsequent treatment to the image after correcting.
3) extract in the right figure correcting rear binocular image and extract target spy title point, centered by each unique point, extract a matrix-block, then the matrix-block that relevance of searches is maximum on the polar curve of left figure.
4) when certain matrix-block maximum correlation on polar curve is greater than certain threshold value, then the center of this matrix-block is the unique point of coupling.By the mathematic interpolation parallax of two pixel coordinates, and calculate the camera coordinates of unique point.
5), after obtaining all unique point camera coordinates, unique point camera coordinates on ground can be utilized to demarcate the relation of ground level in camera coordinates system, build world coordinate system on this basis, and calculate the world coordinates of unique point.
6) build gaussian kernel for target, for pedestrian, pedestrian is at floor projection sub-elliptical.It is relatively more reasonable that elliptical center represents target location, and the central feature point corresponding row people crown, therefore distance ground is also higher, and namely the weights of position are with highly becoming positive correlation.
7) estimate based on Gaussian Kernel Density, utilize mean ?shift algorithm hill climbing process carried out to unique point find probability space local maximum, and cluster is carried out to characteristic point position after climbing the mountain, each clarification of objective point is gathered into a classification, obtains the position at target place.
8) Kalman filtering algorithm is used, suppose in pedestrian's short time it is linear uniform motion, by the tracking results estimated position of previous frame as estimated value, using the cluster result of present frame as measured value, finally estimate that an optimal value is as present frame target position.
Adopt oval kernel function as the position kernel function of tracking pedestrians in the present embodiment, major axis parameter alpha minor axis parameter beta.
As shown in Figure 3, be one of the application scenarios of the present embodiment, the one group of experiment namely carried out on doorway, indoor outward, light can strengthen gradually, changes obvious.For target, color characteristic change obviously, adds the difficulty of target following.Native system uses Short baseline biocular systems, and between two cameras, baseline is shorter, extracts image harris feature, the impact that illumination variation of cancelling out each other during characteristic matching produces.
Have 4 video in windows in each frame in Fig. 3, above two windows represent left and right two figure, the lower left corner represents that target signature point is at floor projection figure, and the lower right corner is 3D effect figure, and each cylinder represents a target.Result shows, this system has stronger adaptability to multiple target tracking under illumination variation sight.
As shown in Figure 4, for the present invention is applied to outdoor optical according to the strong scene producing shade, and target occlusion is more serious.Because shadow character point height is zero, first native system calculates unique point world coordinates, and then filtration is highly the unique point of zero, can well solve the interference that shade brings.Utilize mean-shift Non-surveillance clustering method to carry out cluster to three-dimensional information unique point, and use Kalman filtering algorithm to follow the tracks of, effectively can solve and block, the problem such as crowded.Result shows, and this system has good robustness to shade, the problem such as to block.
Claims (9)
1., based on a multi-object tracking method for kernel function Non-surveillance clustering, comprise the following steps:
1) first obtain the left and right sequence chart of synchronization as input by binocular camera, utilize binocular camera parameter to image rectification;
2) by extracting image characteristic point and matching characteristic, parallax is calculated;
3) utilize the relative camera coordinates position of disparity computation target signature point obtained, i.e. camera coordinates, and complete ground demarcation, thus according to unique point distance floor level, ground shadow character point can be filtered, eliminate the interference of ground area shading;
4) for three-dimensional coordinate unique point, be combined kernel function, Non-surveillance clustering is carried out to the target of uncertain classification number, all unique points of a target are aggregated into a set, the i.e. position of the corresponding observed reading of each classification and direction, and obtain target place present frame in conjunction with the position of previous frame target and direction prediction, the i.e. position of predicted value and direction prediction value, use optimal estimation algorithm finally to obtain position and the direction of optimal objective, thus reach the effect that multiple goal follows the tracks of fast.
2. method according to claim 1, is characterized in that, described image rectification refers to the inside and outside parameter utilizing binocular camera, to the correct image got, makes:
I) one, synchronization space point meets Epipolar geometry theorem in left and right sequence chart;
Ii) determine that a corresponding locus is on the polar curve of another sequence chart for one in the arbitrary width in left and right sequence chart.
3. method according to claim 1, is characterized in that, described matching characteristic refers to extract minutiae in the arbitrary width first in left and right sequence chart, and utilizes the impact point polar curve of matching algorithm in another sequence chart being searched for correspondence.
4. method according to claim 1, it is characterized in that, described ground is demarcated and is referred to: make marks on the ground, extract the unique point on this mark in left and right sequence chart, calculate its camera coordinates, and utilize least square method to ground unique point camera coordinates matching ground level parametric equation.
5. method according to claim 1, is characterized in that, described step 4 specifically comprises:
A) all unique points are filtered, filter out and highly may belong to clarification of objective point, then be each unique point initialization angle θ
0, the anglec of rotation of certain target may be belonged to as each unique point, i.e. target direction;
B) construct the weighted value that kernel density function is each sample point estimated probability, find target location in the corresponding space of local maximum at probability space; The probability space point that during search, constantly search is larger in its contiguous range from a starting point, until reach the position of local maximum point, namely mean ?shift hill climbing process, and use Non-surveillance clustering method to carry out cluster to the characteristic point position after climbing the mountain, classification number represents target number;
C) utilize optimal estimation algorithm, the motion vector new according to the motion vector prediction in target past, upper frame position basis obtains the predicted position of present frame, reaches the effect to the correction of current location measured value.
6. method according to claim 5, is characterized in that, described structure kernel density function refers to, by characteristic point position variable and angle variables, constructs the probability density of this unique point of function representation at floor projection:
Wherein: the local maximum point in probability space E (x, θ) correspond to the densest position of discrete space mid point, the target that present frame detects, H is represented
θ(d
i(θ)), H
x(d
j(x), θ
i) be the kernel function of direction and position, d
i(θ), d
jx () is the normalized cumulant measure function of direction and position, d
i(θ)=|| A
θ(θ-θ
i) ||
2, d
j(θ)=|| A
x(x-x
j) ||
2, the variable x in measure function
jrefer to jth unique point projection coordinate on the ground, variable θ
iat x
jrepresent i-th direction value in the coordinate [0,2 π] of a jth subpoint, w
jrepresent the weighted value of a jth subpoint.
7. method according to claim 5, it is characterized in that, in step b, use Non-surveillance clustering method to carry out cluster to the characteristic point position after climbing the mountain to refer to, being gathered by point very little for distance after climbing the mountain is a class, represent a target, wherein: the distance between two unique points can be expressed as d (x
i, x
j), be then divided into a class when distance is less than certain value.
8. method according to claim 5, is characterized in that, the cluster described in step b, is judged by the variance based on direction distance definition intra-cluster point distance, and the classification in cluster result is the need of partition or merging further.
9. one kind realizes the system of method described in above-mentioned arbitrary claim, it is characterized in that, comprise: camera calibration module, parameter input module, image processing module, feature extracting and matching module, ground demarcating module, ordinate transform module, kernel function cluster module, filter tracking module, showing interface module, wherein: camera calibration module is demarcated the camera parameter of binocular camera and is stored in parameter input module, parameter input module reads the camera parameter demarcated and exports image processing module to, image processing module to correct binocular sequence of pictures according to camera parameter and exports feature extracting and matching module to, feature extracting and matching module extracts target signature from the picture after correction, also ordinate transform module and ground demarcating module is exported respectively to by relevant matches unique point calculating unique point parallax and camera coordinates, ground demarcating module is demarcated ground parameter, and export ground parameter to ordinate transform module when following the tracks of, the camera coordinates of unique point is converted to world coordinates and exports kernel function cluster module to by ordinate transform module carries out cluster, each target is polymerized to a set by kernel function cluster module, namely present frame testing result exports filter tracking module to, filter tracking module estimates optimum target location and direction according to previous frame result and present frame testing result, and show tracking results by showing interface module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510386915.5A CN105160649A (en) | 2015-06-30 | 2015-06-30 | Multi-target tracking method and system based on kernel function unsupervised clustering |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510386915.5A CN105160649A (en) | 2015-06-30 | 2015-06-30 | Multi-target tracking method and system based on kernel function unsupervised clustering |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105160649A true CN105160649A (en) | 2015-12-16 |
Family
ID=54801491
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510386915.5A Pending CN105160649A (en) | 2015-06-30 | 2015-06-30 | Multi-target tracking method and system based on kernel function unsupervised clustering |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105160649A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106203429A (en) * | 2016-07-06 | 2016-12-07 | 西北工业大学 | Based on the shelter target detection method under binocular stereo vision complex background |
| CN107909604A (en) * | 2017-11-07 | 2018-04-13 | 武汉科技大学 | Dynamic object movement locus recognition methods based on binocular vision |
| CN108389228A (en) * | 2018-03-12 | 2018-08-10 | 海信集团有限公司 | Ground detection method, apparatus and equipment |
| CN108537250A (en) * | 2018-03-16 | 2018-09-14 | 新智认知数据服务有限公司 | A kind of target following model building method and device |
| CN110287936A (en) * | 2019-07-02 | 2019-09-27 | 北京字节跳动网络技术有限公司 | Image detecting method, device, equipment and storage medium |
| CN110472553A (en) * | 2019-08-12 | 2019-11-19 | 北京易航远智科技有限公司 | Target tracking method, computing device and the medium of image and laser point cloud fusion |
| CN111080676A (en) * | 2019-12-20 | 2020-04-28 | 电子科技大学 | Method for tracking endoscope image sequence feature points through online classification |
| CN111091078A (en) * | 2019-12-03 | 2020-05-01 | 北京华捷艾米科技有限公司 | Object tracking method and related equipment |
| CN111275829A (en) * | 2020-02-19 | 2020-06-12 | 西北大学 | Three-dimensional local feature descriptor acquisition method and feature descriptor matching method |
| CN111292198A (en) * | 2020-01-21 | 2020-06-16 | 北京电子工程总体研究所 | Cluster target establishing method for restraining stability of central point |
| CN112581537A (en) * | 2019-09-29 | 2021-03-30 | 长沙智能驾驶研究院有限公司 | Binocular camera external parameter inspection method and device, computer equipment and storage medium |
| CN112633096A (en) * | 2020-12-14 | 2021-04-09 | 深圳云天励飞技术股份有限公司 | Passenger flow monitoring method and device, electronic equipment and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101344965A (en) * | 2008-09-04 | 2009-01-14 | 上海交通大学 | Tracking system based on binocular camera shooting |
| US20100332425A1 (en) * | 2009-06-30 | 2010-12-30 | Cuneyt Oncel Tuzel | Method for Clustering Samples with Weakly Supervised Kernel Mean Shift Matrices |
| CN103106659A (en) * | 2013-01-28 | 2013-05-15 | 中国科学院上海微***与信息技术研究所 | Open area target detection and tracking method based on binocular vision sparse point matching |
-
2015
- 2015-06-30 CN CN201510386915.5A patent/CN105160649A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101344965A (en) * | 2008-09-04 | 2009-01-14 | 上海交通大学 | Tracking system based on binocular camera shooting |
| US20100332425A1 (en) * | 2009-06-30 | 2010-12-30 | Cuneyt Oncel Tuzel | Method for Clustering Samples with Weakly Supervised Kernel Mean Shift Matrices |
| CN103106659A (en) * | 2013-01-28 | 2013-05-15 | 中国科学院上海微***与信息技术研究所 | Open area target detection and tracking method based on binocular vision sparse point matching |
Non-Patent Citations (1)
| Title |
|---|
| 蔡岭: "基于立体视觉的目标检测与跟踪", 《中国博士学位论文全文数据库信息科技辑》 * |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106203429A (en) * | 2016-07-06 | 2016-12-07 | 西北工业大学 | Based on the shelter target detection method under binocular stereo vision complex background |
| CN107909604A (en) * | 2017-11-07 | 2018-04-13 | 武汉科技大学 | Dynamic object movement locus recognition methods based on binocular vision |
| CN108389228A (en) * | 2018-03-12 | 2018-08-10 | 海信集团有限公司 | Ground detection method, apparatus and equipment |
| CN108389228B (en) * | 2018-03-12 | 2021-04-20 | 海信集团有限公司 | Ground detection method, device and equipment |
| CN108537250A (en) * | 2018-03-16 | 2018-09-14 | 新智认知数据服务有限公司 | A kind of target following model building method and device |
| CN108537250B (en) * | 2018-03-16 | 2022-06-14 | 新智认知数据服务有限公司 | Target tracking model construction method and device |
| CN110287936A (en) * | 2019-07-02 | 2019-09-27 | 北京字节跳动网络技术有限公司 | Image detecting method, device, equipment and storage medium |
| CN110287936B (en) * | 2019-07-02 | 2021-08-10 | 北京字节跳动网络技术有限公司 | Image detection method, device, equipment and storage medium |
| CN110472553A (en) * | 2019-08-12 | 2019-11-19 | 北京易航远智科技有限公司 | Target tracking method, computing device and the medium of image and laser point cloud fusion |
| CN110472553B (en) * | 2019-08-12 | 2022-03-11 | 北京易航远智科技有限公司 | Target tracking method, computing device and medium for fusion of image and laser point cloud |
| CN112581537A (en) * | 2019-09-29 | 2021-03-30 | 长沙智能驾驶研究院有限公司 | Binocular camera external parameter inspection method and device, computer equipment and storage medium |
| CN112581537B (en) * | 2019-09-29 | 2024-04-19 | 长沙智能驾驶研究院有限公司 | Binocular camera external parameter checking method, binocular camera external parameter checking device, computer equipment and storage medium |
| CN111091078A (en) * | 2019-12-03 | 2020-05-01 | 北京华捷艾米科技有限公司 | Object tracking method and related equipment |
| CN111091078B (en) * | 2019-12-03 | 2023-10-24 | 北京华捷艾米科技有限公司 | Object tracking method and related equipment |
| CN111080676A (en) * | 2019-12-20 | 2020-04-28 | 电子科技大学 | Method for tracking endoscope image sequence feature points through online classification |
| CN111292198A (en) * | 2020-01-21 | 2020-06-16 | 北京电子工程总体研究所 | Cluster target establishing method for restraining stability of central point |
| CN111292198B (en) * | 2020-01-21 | 2022-11-22 | 北京电子工程总体研究所 | Cluster target establishing method for restraining stability of central point |
| CN111275829A (en) * | 2020-02-19 | 2020-06-12 | 西北大学 | Three-dimensional local feature descriptor acquisition method and feature descriptor matching method |
| CN111275829B (en) * | 2020-02-19 | 2022-04-19 | 西北大学 | Three-dimensional local feature descriptor acquisition method and feature descriptor matching method |
| CN112633096A (en) * | 2020-12-14 | 2021-04-09 | 深圳云天励飞技术股份有限公司 | Passenger flow monitoring method and device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105160649A (en) | Multi-target tracking method and system based on kernel function unsupervised clustering | |
| CN104200495B (en) | A kind of multi-object tracking method in video monitoring | |
| CN101701818B (en) | Method for detecting long-distance barrier | |
| CN113506318B (en) | Three-dimensional target perception method under vehicle-mounted edge scene | |
| CN104978567B (en) | Vehicle checking method based on scene classification | |
| Xu et al. | Automatic building rooftop extraction from aerial images via hierarchical RGB-D priors | |
| CN105528794A (en) | Moving object detection method based on Gaussian mixture model and superpixel segmentation | |
| CN103824070A (en) | Rapid pedestrian detection method based on computer vision | |
| CN113223045B (en) | Vision and IMU sensor fusion positioning system based on dynamic object semantic segmentation | |
| Wang et al. | An overview of 3d object detection | |
| CN103593679A (en) | Visual human-hand tracking method based on online machine learning | |
| CN115376109B (en) | Obstacle detection method, obstacle detection device, and storage medium | |
| CN115308732A (en) | Multi-target detection and tracking method integrating millimeter wave radar and depth vision | |
| CN116978009A (en) | Dynamic object filtering method based on 4D millimeter wave radar | |
| CN112432653B (en) | Monocular vision inertial odometer method based on dotted line characteristics | |
| CN116862832A (en) | Three-dimensional live-action model-based operator positioning method | |
| CN117036404A (en) | Monocular thermal imaging simultaneous positioning and mapping method and system | |
| CN114283199B (en) | Dynamic scene-oriented dotted line fusion semantic SLAM method | |
| CN106558065A (en) | The real-time vision tracking to target is realized based on color of image and texture analysiss | |
| Chai et al. | 3D gesture recognition method based on faster R-CNN network | |
| Fardi et al. | Motion-based pedesvtrian recognition from a moving vehicle | |
| Lachachi et al. | LIDAR and Stereo-Camera fusion for reliable Road Extraction | |
| CN113628251A (en) | Smart hotel terminal monitoring method | |
| Tian et al. | Object Tracking Algorithm based on Improved Siamese Convolutional Networks Combined with Deep Contour Extraction and Object Detection Under Airborne Platform. | |
| Li et al. | LIDAR-incorporated traffic sign detection from video log images of mobile mapping system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20151216 |
|
| RJ01 | Rejection of invention patent application after publication |