CN103971380B - Pedestrian based on RGB-D trails detection method - Google Patents
Pedestrian based on RGB-D trails detection method Download PDFInfo
- Publication number
- CN103971380B CN103971380B CN201410186168.6A CN201410186168A CN103971380B CN 103971380 B CN103971380 B CN 103971380B CN 201410186168 A CN201410186168 A CN 201410186168A CN 103971380 B CN103971380 B CN 103971380B
- Authority
- CN
- China
- Prior art keywords
- image
- depth
- target
- depth image
- rgb
- 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.)
- Expired - Fee Related
Links
Abstract
A kind of pedestrian based on RGB D trails detection method.The method first passes through the depth image in monitoring region and detects foreground target, then number of people target is accurately detected according to the depth profile information of hair color information and head and shoulder and positions, finally use camshift algorithm that head part's target is tracked, and then determine whether occur trailing phenomenon.This method is possible not only to effectively identify pedestrian and luggage, pedestrian is carried out accurate metering, will not miss and pedestrian be treated as luggage or by luggage as pedestrian, and the phenomenon trailing many people has higher accuracy of detection, is particularly well-suited to the places such as gate inhibition, security check, station, company's Vomitory.
Description
Technical field
The invention belongs to intelligent monitoring technology field, particularly relate to a kind of pedestrian based on RGB-D and trail detection method.
Background technology
Along with the fast development of computer vision technique, intelligent monitoring has in fields such as security protection, safety check and traffic administrations
And be widely applied, such as number-plate number inspection, pedestrian or the vehicular traffic field such as detection and visualization gate inhibition in violation of rules and regulations.Intelligence
Change monitoring and be possible not only to reduce the cost of manpower monitoring, but also be an important embodiment of urban civilization.
We were frequently seen such as subway gate and crushed the report of the accidents such as passenger in the past, and trace it to its cause the most existing system
System cannot be distinguished by pedestrian and luggage, and has a few people to pass through after cannot be distinguished by swiping the card, if passenger pushes luggage and leads to after swiping the card
Cross gate, the most probably crushed, because gate can miss luggage as pedestrian.Additionally in the porch of some companies all
It is provided with fingerprint or the security check passage such as check card, but often occur a people to check card situation that many people trail.It addition, present stage is for people
The intelligent monitor system that health check-up is surveyed is many based on two dimensional image, therefore, it is difficult to solve that illumination variation, color of object be close and shade etc.
Interference.
Chinese invention patent application the 201210240640.0th discloses a kind of pedestrian only using depth information examine
Survey method, the method is that adult, child and luggage are made a distinction by the difference with depth threshold, and shortcoming is with little to heavy luggage
The differentiation of child is the most difficult.Chinese invention patent application the 201110026465.0th disclose a kind of based on depth information
Pedestrian detection method, the method is by a large amount of depth images extraction feature is built disaggregated model, then will be from depth map
The feature extracted in Xiang is input in disaggregated model, to judge whether comprise human body in region, to realize human detection.But the party
The shortcoming of method is that computation complexity is high, and cannot be carried out the tracing detection of human body target.
Summary of the invention
In order to solve the problems referred to above, it is an object of the invention to provide one and be possible not only to effectively identify pedestrian and row
Lee, carries out accurate metering, and the pedestrian based on RGB-D that the phenomenon trailing many people has relatively high measurement accuracy trails pedestrian
Detection method.
In order to achieve the above object, the pedestrian based on RGB-D that the present invention provides trails detection method and includes entering in order
The following step of row:
1) colour imagery shot is disposed side by side on monitoring with depth camera in the way of monitoring viewing angles-both vertical is downward by foundation
The surface of passage, the prison that colour imagery shot is perpendicular with monitor channel length direction with the straight line at depth camera place simultaneously
Control scene, and utilize colour imagery shot and the RGB image in depth camera Real-time Collection monitor channel and depth image;
2) use background subtraction method based on depth image to carry out the foreground extraction of depth image, obtain foreground depth image
With prospect RGB image;
3) utilize above-mentioned foreground depth image and prospect RGB image that number of people target is detected, determine head zone;
4) in above-mentioned head zone, carry out number of people target following, thus judge whether to trail phenomenon;
In step 2) in, the described foreground extraction using background subtraction method based on depth image to carry out depth image,
The method obtaining foreground depth image and prospect RGB image is:
If background depth image is Bd(x, y), the depth image of present frame is Id(x, y), RGB image be Ic(x, y), only
Need directly to the two do difference can try to achieve foreground mask image M (x, y) be:
Fd(x, y)=Id(x,y)·M(x,y)
Fc(x, y)=Ic(x,y)·M(x,y)
In formula, MtFor decision threshold, (x y) is foreground image to F.Fd(x y) is foreground depth image, Fc(x y) is prospect
RGB image.
In step 3) in, number of people target is detected by described Utilization prospects depth image and prospect RGB image, determines
The method of head zone is: first according to the cluster feature of hair color, extracts candidate's head from above-mentioned prospect RGB image
Region, portion, and by candidate head area maps to foreground depth image, the then foreground depth image to candidate head region
Carry out rim detection, and use hough change detection class therein circle ring area, remove pseudo-header area finally according to priori
Territory, obtains contouring head figure.
In step 4) in, described carries out number of people target following in head zone, thus judges whether to trail now
The method of elephant is: first using the boundary rectangle of the head zone of above-mentioned acquisition as initial ranging frame, to depth value in rectangle frame
Add up, obtain the degree of depth histogram model of head target;Then by the degree of depth Histogram backprojection of target to being followed the tracks of
In the depth image of frame, obtain the degree of depth probability distribution graph of head target, in this probability distribution graph, the pixel table that probability is big
This pixel bright is that the probability of target is bigger;Finally according to the tracking result to head target, and connected applications scene is sentenced
Break and whether trail phenomenon.
The pedestrian based on RGB-D that the present invention provides trails detection method and first passes through the depth image detection in monitoring region
Go out foreground target, then according to the depth profile information of hair color information and head, number of people target is accurately detected and
Location, finally uses camshift algorithm to be tracked head part's target, and then determines whether occur trailing phenomenon.This method
It is possible not only to effectively identify pedestrian and luggage, pedestrian is carried out accurate metering, will not miss and pedestrian as luggage or will be gone
Li Dangcheng pedestrian, and the phenomenon trailing many people has higher accuracy of detection, is particularly well-suited to gate inhibition, security check, station, public affairs
The places such as department's Vomitory.
Accompanying drawing explanation
The pedestrian based on RGB-D that Fig. 1 provides for the present invention trails the monitoring scene top view employed in detection method.
The pedestrian based on RGB-D that Fig. 2 present invention provides trails people's head inspecting method flow chart in detection method.
Fig. 3 trails the number of people in detection method for the pedestrian based on RGB-D that the present invention provides and follows the tracks of and trail method flow
Figure.
Detailed description of the invention
The pedestrian based on RGB-D provided this utility model with specific embodiment below in conjunction with the accompanying drawings trails detection method
It is described in detail.
As shown in Fig. 1 Fig. 3, the pedestrian based on RGB-D that the present invention provides trails detection method and includes carrying out in order
The following step:
1) set up as shown in Figure 1 by colour imagery shot 1 and depth camera 2 by monitor viewing angles-both vertical downward in the way of also
Row is arranged on the surface of monitor channel 3, and colour imagery shot 1 should be the most close with the distance of depth camera, to reduce
The interference caused because of different angulars field of view;The straight line at colour imagery shot 1 and depth camera 2 place and monitor channel simultaneously
The monitoring scene that 3 length directions are perpendicular, and utilize in colour imagery shot 1 and depth camera 2 Real-time Collection monitor channel 3
RGB image and depth image;
Original monitor video is only provided that rgb video image, but but lost spatial information in image, and we can only
From RGB color, go to distinguish different targets by some feature, but in space, the segmentation of target should be empty with it
Between connectivity carry out splitting.It is accurate that the present invention combines that RGB color information and spatial depth information complete pedestrian target
Monitoring.Because the interference factors such as being obtained without of depth image is close by target and background color, illumination variation and shade
Impact, it is possible to degree of precision extract foreground target.Because of eclipse phenomena is inevitable in monitoring field but serious shadow
Ringing monitoring result, in order to reduce the eclipse phenomena between moving target as far as possible, this method uses method shown in Fig. 1, by colour
Photographic head 1 and depth camera 2 are arranged on above passage and are monitored with visual angle vertically downward.
2) use background subtraction method based on depth image to carry out the foreground extraction of depth image, obtain foreground depth image
With prospect RGB image;
Background subtraction method is one of conventional means of foreground detection, and it is by comparing conventional images and known Background
Picture, detects that the bigger place of difference is defined as prospect.Mostly current method is video flowing based on RGB, and when by illumination, the moon
Shadow and when background colour and foreground target color are close when affecting, can seriously reduce the accuracy of foreground detection.The present invention
Use background subtraction method based on depth image.Because depth image reaction is the spatial information of scene, every in image
Individual pixel value represents corresponding object and arrives the distance of photographic head plane.Depth image will not by shade in scene, illumination with
And the impact of color change.Concrete grammar is as follows:
If background depth image is Bd(x, y), the depth image of present frame is Id(x, y), RGB image be Ic(x, y), only
Need directly to the two do difference can try to achieve foreground mask image M (x, y) be:
Fd(x, y)=Id(x,y)·M(x,y)
Fc(x, y)=Ic(x,y)·M(x,y)
In formula, MtFor decision threshold, (x y) is foreground image to F.Fd(x y) is foreground depth image, Fc(x y) is prospect
RGB image.
3) utilize above-mentioned foreground depth image and prospect RGB image that number of people target is detected, determine head zone;
In the case of photographic head vertically sets up, the number of people is detected frequently with based on hair color and contouring head
Method.Because Aisan's hair color is relatively deep, according to statistics, in RGB color, the R component of hair color is gathered in 0-35
Between.Typically can be by this color cluster feature extraction head candidate region of hair.It addition, the number of people in monitored picture
Contour approximation is circular, and a lot of scholars by judging the exact position of head to the detection of annulus.But traditional contours extract
Method is that the gradient according to gradation of image is carried out.When hair color and background (such as clothes, monitoring scene) color are close,
Just it is difficult to extract exactly the profile of the number of people.
Physical segmentation in view of target is built upon above the discontinuity of its degree of depth, and the number of people is in human body
Eminence, head and shoulder is substantially without being blocked, so its contour feature is the most prominent, substantially distinguishes over other healths from depth information
Position, therefore can distinguish number of people target from image exactly.
Therefore the present invention proposes a kind of number of people profile testing method based on depth image, first according to hair color
Cluster feature, extracts candidate head region from above-mentioned prospect RGB image, and by candidate head area maps to foreground depth
In image, then the foreground depth image to candidate head region carries out rim detection, and uses hough change detection therein
Class circle ring area, removes pseudo-head zone finally according to priori, obtains contouring head figure.Concrete grammar is as follows:
According to hair color cluster feature, set threshold range R ∈ [0,35] of hair R component, from prospect RGB image
Extract candidate head region.
In formula, FR(x y) is prospect RGB image Fc(x, R channel components y), Hm(x y) is candidate head region mask.
1, the boundary rectangle of candidate head region mask is asked for, and by this rectangle centered by diagonal intersection point, by length and width
3 pixel wide of extension respectively, to ensure that rectangle frame can be completely covered the actual physics profile of head.Then this rectangle frame is reflected
It is mapped to foreground depth image Fd(x, y) in, obtain the depth image in candidate head region, be set to H hered(x,y)。
2, the canny operator depth image H to candidate head region is usedd(x, y) carries out rim detection, by arranging conjunction
Suitable detection threshold value, can obtain the edge contour image of refinement.
3, the annulus in hough change detection edge contour image is used.
4, remove pseudo-head zone according to priori, obtain head zone.
Owing to contouring head is not standard circular profile, but the class circle contour of sub-circular, same head exists
There may be multiple output after hough conversion, according to prioris such as photographic head height and number of people sizes, use hough
When conversion carries out annulus detection, first set round least radius as Rmin, maximum radius is RmaxAnd the center of circle of adjacent rings
Distance RL>Rmax.It addition, the present invention extracts the depth value of head zone from multi-amplitude deepness image, ask for its average M.Statistics knot
Fruit shows: in the depth image of whole head zone, in addition to the 0 value region that subregion occurs because detection is inaccurate, remaining
The depth value of part is basically stable in the range of M ± 5cm.After obtaining circle ring area by hough, can be by annulus half
Footpath Ri(i is annulus sequence number) tries to achieve its area Si, then ask for maximum depth value D in circle ring area, and with [D-10cm, D] for model
Enclose extraction mask, statistical mask region area S from circle ring areami, work as q=Smi/Si>q0Time (q0For setting threshold value), can be
Judge that this annulus is as head zone eventually.
4) in above-mentioned head zone, carry out number of people target following, thus judge whether to trail phenomenon;
In terms of target following, camshift algorithm is really a kind of method of real-time high-efficiency, and it utilizes the color in region
Information, completes the tracking to moving target by the way of cluster.Traditional camshift algorithm must be by image from RGB face
Color space transformation is to hsv color space, and then the rectangular histogram of recycling H component sets up the color probability model of target.But when fortune
Moving-target and background color close to time, its testing result is difficult to satisfactory.
The present invention is inspired from the camshift algorithm that motion target tracking field is commonly used, because vertical at photographic head
In the case of erection, pedestrian pass by monitored space time, head part distance photographic head plane distance be almost unchanged, i.e. in the degree of depth
The depth value scope for people's head region that embodies on image is almost unchanged.Therefore the present invention proposes a kind of based on depth image
Camshift target tracking method, the H component used in tradition camshift algorithm is replaced with depth component by the method, and then
Camshift algorithm is utilized to carry out target following.First using the boundary rectangle of the head zone of above-mentioned acquisition as initial ranging
Frame, adds up depth value in rectangle frame, obtains the degree of depth histogram model of head target;Then by the degree of depth Nogata of target
Figure back projection, in the depth image of institute's tracking frame, obtains the degree of depth probability distribution graph of head target, in this probability distribution graph
In, the pixel that probability is big shows that the probability that this pixel is target is bigger;Finally according to the tracking result to head target,
And connected applications scene judges whether to trail phenomenon.
Concrete grammar is as follows:
1, the minimum enclosed rectangle of head zone is taken as search window, if there being multiple head zone respectively they to be marked
Note.Ask for the depth information rectangular histogram of depth image in search box;
2, by Histogram backprojection to former depth image, degree of depth probability distribution graph is obtained;
3, application camshift algorithm, finds the region close with search window in degree of depth probability distribution graph and is target area
Whether territory, thus determine and occur trailing.
Claims (3)
1. pedestrian based on RGB-D trails a detection method, and it includes the following step carried out in order:
1) set up by colour imagery shot (1) and depth camera (2) by monitor viewing angles-both vertical downward in the way of be disposed side by side on prison
The surface of control passage (3), the simultaneously straight line at colour imagery shot (1) and depth camera (2) place and monitor channel (3) length
The monitoring scene that direction is perpendicular, and utilize colour imagery shot (1) and depth camera (2) Real-time Collection monitor channel (3)
In RGB image and depth image;
2) use background subtraction method based on depth image to carry out the foreground extraction of depth image, obtain foreground depth image and front
Scape RGB image;
3) utilize above-mentioned foreground depth image and prospect RGB image that number of people target is detected, determine head zone;
4) in above-mentioned head zone, carry out number of people target following, thus judge whether to trail phenomenon;
It is characterized in that: in step 2) in, before described employing background subtraction method based on depth image carries out depth image
Scape extracts, and the method obtaining foreground depth image and prospect RGB image is:
If background depth image is Bd(x, y), the depth image of present frame is Id(x, y), RGB image be Ic(x y), only needs straight
Connect Id(x,y)、Bd(x, y) do difference can try to achieve foreground mask image M (x, y) be:
Fd(x, y)=Id(x,y)·M(x,y)
Fc(x, y)=Ic(x,y)·M(x,y)
In formula, MtFor decision threshold, (x y) is foreground image to F;Fd(x y) is foreground depth image, Fc(x y) is prospect RGB
Image.
Pedestrian based on RGB-D the most according to claim 1 trails detection method, it is characterised in that: in step 3) in, institute
Number of people target is detected by Utilization prospects depth image and the prospect RGB image stated, determines that the method for head zone is: first
According to the cluster feature of hair color, from above-mentioned prospect RGB image, extract candidate head region, and by candidate head region
It is mapped in foreground depth image, then the foreground depth image in candidate head region is carried out rim detection, and use hough
Change detection class therein circle ring area, removes pseudo-head zone finally according to priori, obtains contouring head figure.
Pedestrian based on RGB-D the most according to claim 1 trails detection method, it is characterised in that: in step 4) in, institute
That states carries out number of people target following in head zone, thus judges whether that the method trailing phenomenon is: first by above-mentioned
Depth value in rectangle frame, as initial ranging frame, is added up, is obtained head target by the boundary rectangle of the head zone obtained
Degree of depth histogram model;Then by the degree of depth Histogram backprojection of target to the depth image of institute's tracking frame, obtain to the end
The degree of depth probability distribution graph of portion's target, in this probability distribution graph, the pixel that probability is big show this pixel be target can
Energy property is bigger;Finally according to the tracking result to head target, and connected applications scene judges whether to trail phenomenon.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410186168.6A CN103971380B (en) | 2014-05-05 | 2014-05-05 | Pedestrian based on RGB-D trails detection method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410186168.6A CN103971380B (en) | 2014-05-05 | 2014-05-05 | Pedestrian based on RGB-D trails detection method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103971380A CN103971380A (en) | 2014-08-06 |
CN103971380B true CN103971380B (en) | 2016-09-28 |
Family
ID=51240829
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410186168.6A Expired - Fee Related CN103971380B (en) | 2014-05-05 | 2014-05-05 | Pedestrian based on RGB-D trails detection method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103971380B (en) |
Families Citing this family (23)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104657993B (en) * | 2015-02-12 | 2018-04-17 | 北京格灵深瞳信息技术有限公司 | A kind of camera lens occlusion detection method and device |
CN104751491B (en) * | 2015-04-10 | 2018-01-23 | 中国科学院宁波材料技术与工程研究所 | A kind of crowd's tracking and people flow rate statistical method and device |
CN105225230B (en) * | 2015-09-11 | 2018-07-13 | 浙江宇视科技有限公司 | A kind of method and device of identification foreground target object |
CN105741271B (en) * | 2016-01-25 | 2021-11-16 | 上海物联网有限公司 | Method for detecting object in depth image |
CN106096512B (en) * | 2016-05-31 | 2020-08-25 | 上海美迪索科电子科技有限公司 | Detection device and method for recognizing vehicle or pedestrian by using depth camera |
CN107491712A (en) * | 2016-06-09 | 2017-12-19 | 北京雷动云合智能技术有限公司 | A kind of human body recognition method based on RGB D images |
CN106384353A (en) * | 2016-09-12 | 2017-02-08 | 佛山市南海区广工大数控装备协同创新研究院 | Target positioning method based on RGBD |
CN106454229A (en) * | 2016-09-27 | 2017-02-22 | 成都理想境界科技有限公司 | Monitoring method, camera device, image processing device and monitoring system |
CN106530310B (en) * | 2016-10-25 | 2018-02-23 | 深圳大学 | A kind of pedestrian count method and device based on the identification of human body overhead |
CN106778655B (en) * | 2016-12-27 | 2020-05-05 | 华侨大学 | Human body skeleton-based entrance trailing entry detection method |
CN108335308A (en) * | 2017-01-20 | 2018-07-27 | 深圳市祈飞科技有限公司 | A kind of orange automatic testing method, system and intelligent robot retail terminal |
CN107230226A (en) * | 2017-05-15 | 2017-10-03 | 深圳奥比中光科技有限公司 | Determination methods, device and the storage device of human body incidence relation |
CN107221058A (en) * | 2017-05-25 | 2017-09-29 | 刘萍 | Intelligent channel barrier system |
CN108564063B (en) * | 2018-04-27 | 2020-12-18 | 北京华捷艾米科技有限公司 | Palm positioning method and system based on depth information |
CN111192391B (en) * | 2018-10-25 | 2022-09-23 | 杭州海康威视数字技术股份有限公司 | Pedestrian passageway gate control method and device based on images and/or videos |
CN109688452A (en) * | 2018-12-04 | 2019-04-26 | 深圳市子瑜杰恩科技有限公司 | Pagination Display stage property stacking method and Related product |
CN109977109B (en) * | 2019-04-03 | 2021-04-27 | 深圳市甲易科技有限公司 | Track data accompanying analysis method |
CN110516602A (en) * | 2019-08-28 | 2019-11-29 | 杭州律橙电子科技有限公司 | A kind of public traffice passenger flow statistical method based on monocular camera and depth learning technology |
CN111144231B (en) * | 2019-12-09 | 2022-04-15 | 深圳市鸿逸达科技有限公司 | Self-service channel anti-trailing detection method and system based on depth image |
CN110969747A (en) * | 2019-12-11 | 2020-04-07 | 盛视科技股份有限公司 | Anti-following access control system and anti-following method |
CN111723770B (en) * | 2020-06-30 | 2020-12-18 | 四川兴事发门窗有限责任公司 | Anti-trailing gate system and method based on image recognition |
WO2022151507A1 (en) * | 2021-01-18 | 2022-07-21 | 深圳市大疆创新科技有限公司 | Movable platform and method and apparatus for controlling same, and machine-readable storage medium |
CN113065397B (en) * | 2021-03-02 | 2022-12-23 | 南京苏宁软件技术有限公司 | Pedestrian detection method and device |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102737235A (en) * | 2012-06-28 | 2012-10-17 | 中国科学院自动化研究所 | Head posture estimation method based on depth information and color image |
CN102999892A (en) * | 2012-12-03 | 2013-03-27 | 东华大学 | Intelligent fusion method for depth images based on area shades and red green blue (RGB) images |
CN103150559A (en) * | 2013-03-01 | 2013-06-12 | 南京理工大学 | Kinect three-dimensional depth image-based head identification and tracking method |
CN103514429A (en) * | 2012-06-21 | 2014-01-15 | 夏普株式会社 | Method for detecting specific part of object and image processing equipment |
-
2014
- 2014-05-05 CN CN201410186168.6A patent/CN103971380B/en not_active Expired - Fee Related
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103514429A (en) * | 2012-06-21 | 2014-01-15 | 夏普株式会社 | Method for detecting specific part of object and image processing equipment |
CN102737235A (en) * | 2012-06-28 | 2012-10-17 | 中国科学院自动化研究所 | Head posture estimation method based on depth information and color image |
CN102999892A (en) * | 2012-12-03 | 2013-03-27 | 东华大学 | Intelligent fusion method for depth images based on area shades and red green blue (RGB) images |
CN103150559A (en) * | 2013-03-01 | 2013-06-12 | 南京理工大学 | Kinect three-dimensional depth image-based head identification and tracking method |
Non-Patent Citations (2)
Title |
---|
融合Kinect深度和颜色信息的机器人视觉系统研究;吴鑫;《中国优秀硕士学位论文全文数据库 信息科技辑》;20131215(第S2期);第4.1.2节 * |
视频监控中的人体头部检测与跟踪;丁小童;《中国优秀硕士学位论文全文数据库 信息科技辑》;20120715(第07期);第1.2节第3段第3-7行,第5.1节、图4.1,图5.1 * |
Also Published As
Publication number | Publication date |
---|---|
CN103971380A (en) | 2014-08-06 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103971380B (en) | Pedestrian based on RGB-D trails detection method | |
CN102214291B (en) | Method for quickly and accurately detecting and tracking human face based on video sequence | |
TWI409718B (en) | Method of locating license plate of moving vehicle | |
CN103218816B (en) | A kind of crowd density estimation method and people flow rate statistical method based on video analysis | |
CN103324913B (en) | A kind of pedestrian event detection method of Shape-based interpolation characteristic sum trajectory analysis | |
CN101794385B (en) | Multi-angle multi-target fast human face tracking method used in video sequence | |
CN102880863B (en) | Method for positioning license number and face of driver on basis of deformable part model | |
CN103150549B (en) | A kind of road tunnel fire detection method based on the early stage motion feature of smog | |
CN102915433B (en) | Character combination-based license plate positioning and identifying method | |
CN102622584B (en) | Method for detecting mask faces in video monitor | |
CN104978567B (en) | Vehicle checking method based on scene classification | |
CN103810703B (en) | A kind of tunnel based on image procossing video moving object detection method | |
CN105046206B (en) | Based on the pedestrian detection method and device for moving prior information in video | |
CN104504369A (en) | Wearing condition detection method for safety helmets | |
CN103473554A (en) | People flow statistical system and people flow statistical method | |
CN104183142A (en) | Traffic flow statistics method based on image visual processing technology | |
CN103310194A (en) | Method for detecting head and shoulders of pedestrian in video based on overhead pixel gradient direction | |
CN105893962A (en) | Method for counting passenger flow at airport security check counter | |
CN104778727A (en) | Floating car counting method based on video monitoring processing technology | |
US10650249B2 (en) | Method and device for counting pedestrians based on identification of head top of human body | |
CN104123734A (en) | Visible light and infrared detection result integration based moving target detection method | |
CN104463138A (en) | Text positioning method and system based on visual structure attribute | |
CN104573697A (en) | Construction hoist lift car people counting method based on multi-information fusion | |
CN104143077B (en) | Pedestrian target search method and system based on image | |
CN103605983A (en) | Remnant detection and tracking method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20160928 Termination date: 20190505 |
|
CF01 | Termination of patent right due to non-payment of annual fee |