CN103886292B - Night vehicle target stable tracking method based on machine vision - Google Patents
Night vehicle target stable tracking method based on machine vision Download PDFInfo
- Publication number
- CN103886292B CN103886292B CN201410105323.7A CN201410105323A CN103886292B CN 103886292 B CN103886292 B CN 103886292B CN 201410105323 A CN201410105323 A CN 201410105323A CN 103886292 B CN103886292 B CN 103886292B
- Authority
- CN
- China
- Prior art keywords
- tracking
- stl
- vehicle
- ptl
- target
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention provides a night vehicle target stable tracking method based on machine vision. The method includes the steps that through judging the state of a stable tracking queue, original images are sent to a vehicle tracking module or a vehicle detecting module for processing; after the images are received by the vehicle tracking module, targets are tracked through the stable tracking queue; areas which are already successfully tracked by the stable tracking queue are marked in the images, and the marked images are transmitted to the vehicle detecting module; the received images are detected through the vehicle detecting module to obtain detection results; the detection results are screened through a pre-tracking queue, obtained targets hit continuously are taken as stable tracking targets, and the stable tracking targets are added to the stable tracking queue. The night vehicle target stable tracking method is used for a night vehicle-mounted road target detecting system based on vision and improving the robustness of the target detecting and tracking system, and the method can be used in the intelligent vehicle anticollision field and the vehicle tracking technical field.
Description
Technical field
The invention belongs to Vehicular intelligent anti-collision early warning (the Vehicle Intelligent Collision of view-based access control model
Warning) field, particularly to a kind of tenacious tracking method of the road at night time vehicle target based on machine vision.
Background technology
Since nineteen nineties, world community proceeds by intelligent transportation system(ITS)Construction.ITS passes through
Using advanced information technology, people, car, road three are flexibly combined, in ITS, an important subsystem is exactly vehicle
Intelligent anticollision early warning system (Vehicle Intelligent Collision Warning System), vehicle detection module with
And tracking module is the important component part of this system.
Vehicle tracking algorithm currently used for the view-based access control model of onboard system scene can be summarized as three classes:Based on template,
Tracking based on Kalman filtering with based on neighborhood.Feature extraction coupling is largely dependent upon based on the method for template
Algorithm, accuracy is high to template dependant degree.Need image is carried out with traversal detection based on the method for Kalman filtering, take system
System resource is many, calculates the time long.The region that tracking based on neighborhood occurs in target high probability carries out detecting and tracking, and searches
Rope whole image region is compared and amount of calculation can be greatly decreased, and time performance is good, but what presence caused because travelling road conditions difference
Photographic head vibrations are led to follow the tracks of target problem easy to lose, bring tracking target to lose as Fig. 1 (a) institute because of photographic head vibrations
Show.
For the drawbacks described above based on neighborhood tracking, the present invention proposes one kind and adopts Pre-tracking queue and tenacious tracking
Deque's tracking of queue.The present invention is on the basis of based on neighborhood tracking, continuous using Pre-tracking queue screening
The detection target of hit, as following the tracks of target, to strengthen the stability of tracking, is strengthened to tracking using tenacious tracking queue
The fault-tolerance that target is lost, therefore, reduces the image trace mesh that the vehicle-mounted camera vibrations causing because travelling road conditions difference lead to
The situation that mark is lost, thus strengthen the anti-interference of neighborhood image target detection tracking method.The method is applied to view-based access control model
Vehicle mounted road object detection system, strengthen target detection tracking system robustness.
Content of the invention
In the application scenarios followed the tracks of for the road vehicle target detection based on monocular vision night onboard system, traditional
It is vulnerable to vehicle-mounted camera vibrations, flating, easily causes detecting and tracking target to lose based on the wireless vehicle tracking of neighborhood
Problem, the invention discloses a kind of tenacious tracking method of the road at night time vehicle target based on monocular vision, for vehicle-mounted
The stable detection of road at night time target and tracking.
The present invention, on the basis of the tracking based on neighborhood, obtains by using Pre-tracking queuing method and stablizes mesh
Stable objects are tracked in neighborhood by mark using tenacious tracking queuing method, simultaneously also using tracing area labelling
Strategy, can reduce detection and tracing area when being tracked, and improve real-time additionally it is possible to strengthen tracking to photographic head
Vibrations and cause the anti-interference that target loses, and the stability of tracking and accuracy.Invention solves its technical problem
Technical scheme as follows:
Assume that video camera is arranged in the middle of vehicle front windshield and towards dead ahead, the i.e. angle of pitch of video camera, course deviation
Angle, the anglec of rotation are zero.
Step1:Receive original image.Vehicle tracking module receives a frame original image srcImg input from photographic head,
Judge whether tenacious tracking queue is empty, if non-NULL, by vehicle tracking resume module srcImg, skip to Step2;If
Sky, then transfer to vehicle detection module directly to process srcImg, skip to Step4.
Step2:Follow the tracks of target using the tenacious tracking queuing method based on neighborhood.Using tenacious tracking queue, stable
In the neighborhood of tracking queue element, srcImg is tracked, obtains tracking result, and update each element of tenacious tracking queue.
Step3:Labelling tracing area.Letter using the tenacious tracking queue each element being updated in previous step
Breath, to the labelling having been followed the tracks of target area in srcImg.
Step4:Different images that vehicle detection module receives to it and carry out heavy duty, then carried out by vehicle detection module
Detection obtains testing result.
Step5:Using Pre-tracking queuing method selective mechanisms result.Obtain vehicle detection result from previous step, make
Screened with Pre-tracking queue, thus the tenacious tracking target continuously hit, finally using tenacious tracking target as new
Element add to tenacious tracking queue, then skip to Step1, until photographic head no longer inputs new image.
The invention has the advantages that:
1st, the detection target that Pre-tracking queuing method screening continuous hit is used, as following the tracks of target, decreases vehicle detection
The negative effect to tracking result for the partial error testing result of module, enhances the stability of tracking result.
The element the 2nd, with the tenacious tracking queue of life cycle has certain fault-tolerance to following the tracks of target and losing, thus
The tracking target causing because of photographic head vibrations can be reduced lose, improve the anti-interference of tracking
3rd, using the strategy of tracing area labelling, by the tracking result of tenacious tracking queue come labelling tracking area
Domain, then using this image as the input of vehicle detection module so that required detection zone minimizing during vehicle detection, save
Onboard system computing resource, improves the speed of service.
Brief description
Fig. 1 (a) is to cause because of vehicle vibration to follow the tracks of the exemplary plot that target is lost.
Fig. 1 (b) is that the tenacious tracking queuing method of the present invention enhances the exemplary plot of the anti-interference to vehicle vibration.
Fig. 2 is the wireless vehicle tracking flow chart that the present invention is used for onboard system.
Fig. 3 (a) is the exemplary plot having flase drop result in vehicle detection result.
Fig. 3 (b) is that the Pre-tracking queuing method of the present invention removes the exemplary plot after flase drop.
Specific embodiment
Below in conjunction with the accompanying drawings, specific embodiments of the present invention are described in further detail.
The present invention is mainly used on the basis of the vehicle at night object detection information of existing monocular video image, according to going through
The detecting and tracking result of history video frame image carries out continuous-stable tracing detection to the vehicle target position in subsequent image frames, solution
The image object shake that certainly the taillight target information of night front vehicles causes because of road conditions difference, causes to follow the tracks of easy to lose the asking of target
Topic, strengthens the robustness of road at night time target following, improves the accuracy rate of vehicle-mounted vehicle at night target detection.
The implementation process of the present invention to be described with reference to Fig. 2 execution step:
Step 1 receives original image:
Vehicle tracking module receives a frame original image srcImg input from photographic head, whether judges tenacious tracking queue
For sky, if non-NULL, srcImg is continued with by vehicle tracking module, skips to step 2;If sky, then srcImg is handed over
Directly processed by vehicle detection module, skip to step 4.
The tenacious tracking queuing method based on neighborhood for the step 2:
This step is in the neighborhood stablizing queue element (QE), according to clarification of objective, target is tracked.Vehicle detection
Result set, Pre-tracking queue, each element of tenacious tracking queue, have the set of a feature(Such as vehicle at night car
The area of lamp, depth-width ratio etc.)It is vehicle feature in the picture to describe this element.Here description vehicle in the picture
The set of feature is defined as feature set.If STL represents tenacious tracking queue, STLiI-th element for STL,
For STLiFeature set,For STLiThe region occurring in former frame, enArea is contiguous range coefficient,
Represent STLiThe threshold set used when being detected, α represents the coefficient following the tracks of threshold set,Represent STLiLife Cycle
Phase, MAXLN is the maximum lifetime of STL element, NMSTLNumber of elements for STL.The tenacious tracking queuing method of the present invention
Enhance shown in effect such as Fig. 1 (b) of the anti-interference to vehicle vibration.Tenacious tracking queuing method based on neighborhood is specifically real
Apply step as follows:
Step1:In srcImg* in enArea, if can allow?* α
Whole characteristic matching successes under constraint.If it can, willIn all features in srcImg update, skip to
Step2, if it is not, skip to Step3.
Step2:Update STL using the hit results in Step1iIn srcImgMake simultaneously
Plus one, untilReach MAXLN.Skip to Step4.
Step3:Subtract one, whenDuring equal to zero, by STLiDelete and by NM from STLSTLSubtract one.Skip to
Step4.
Step4:I adds one, if i is more than NMSTLWhen, this algorithm terminates.Otherwise, skip to Step1.
Step 3 labelling tracing area:
Because onboard system application scenarios have tenacious tracking target will not suddenly disappear, that is, continuity features in target
With target simply to the feature of adjacent domain movement.This step is to utilize These characteristics, according to vehicle tracking module at upper one
In step, successfully track the feature set after target update, after extrapolating new tracking mesh target area, then be tagged to
Image transmitting is to vehicle detection module.Algorithm is specific as follows:
Input:The original image of StaTrackList- tenacious tracking queue srcImg- photographic head input
Output:Image after lableImg- labelling
for each StaTrackList[i].vector inStaTrackList doUpdateStaTrackList
[i].areausingStaTrackList[i].vector;
end for
for each StaTrackList[i].area inStaTrackList dolabelImg.label
(StaTrackList[i].area);
end for
If lableImg is the image after labelling tracing area, lableImg.lable represents that the parameter using input will
LableImg labelling, StaTrackList [i] .vector and StaTrackList [i] .area represents tenacious tracking queue respectively
The feature set of i-th element and the region that occurs in former frame of i-th element.After above-mentioned algorithm tag, will
LableImg transmits to vehicle detection module as input picture.
Step 4 vehicle detection module receives input image simultaneously detects:
Different images that vehicle detection module receives to it and carry out heavy duty, then carry out detecting by vehicle detection module
To testing result.If when STL is not space-time in step 1, the input picture that vehicle detection module receives is labelling tracking area
Image lableImg after domain, and STL is space-time, the input picture that vehicle detection module receives is srcImg.Through right
Input picture detection after, obtain detection result sets DR incoming to vehicle tracking module as Pre-tracking queue input.
Step 5 uses Pre-tracking queuing method selective mechanisms result:
This step mainly uses the DR that previous step obtains, and is screened using Pre-tracking queue, thus obtaining continuous
Tenacious tracking target is finally added to tenacious tracking queue by the tenacious tracking target of hit as new element, and in DR
Shown in flase drop result such as Fig. 3 (a).If DRiFor i-th element in DR, NMDRFor the quantity of element in DR, PTL represents Pre-tracking
Queue, PTLjJ-th element for PTL, NMPTLFor the quantity of element in PTL,WithIt is respectively
DRiFeature set and PTLjFeature set,For PTLjContinuous hit-count,For PTLjWhether it is continuous
The labelling of hit, MAXHTM is the continuous hit-count that detection target can become tenacious tracking target.When this step is held
When row finishes, skip to step 1.The Pre-tracking queuing method of the present invention removes shown in the effect such as Fig. 3 (b) after flase drop.Pre- with
Track queuing method specific implementation step is as follows:
Step1:WillWithCarry out the coupling of feature, if can not all the match is successful skips to
Step2, if can all the match is successful; skip to Step3.
Step2:J plus one, if j is more than NMPTL, then skip to Step4, otherwise, skip to Step1.
Step3:WillPlus one,It is entered as true, ifEqual to MAXHTM, then will
PTLjAdd STL as new element and delete from PTL, NMPTLSubtract one.Skip to Step4.
Step4:I adds one, and j is entered as 1, if i is more than NMDRWhen.Skip to Step5.Otherwise, skip to Step1.
Step5:Make j ∈ [1, NMPTL], for any j, ifValue be not true, then by PTLjFrom pre- with
Delete in track queue.Algorithm leaves it at that.
Claims (1)
1. a kind of vehicle at night target tenacious tracking method based on machine vision is it is characterised in that the method includes following step
Suddenly:
Step 1:Receive original image:
Vehicle tracking module receives a frame original image srcImg input from photographic head, judges that whether tenacious tracking queue is
Sky, if non-NULL, by vehicle tracking resume module image srcImg, skips to step 2;If sky, then by image srcImg
Transfer to vehicle detection module directly to process, skip to step 4;
Step 2:Follow the tracks of target using the tenacious tracking queuing method based on neighborhood:
Using tenacious tracking queue, in the neighborhood of tenacious tracking queue element (QE), image srcImg is tracked, is followed the tracks of
As a result, and update each element of tenacious tracking queue, specifically:
If STL represents tenacious tracking queue, STLiI-th element for STL,For STLiFeature set,For STLiThe region occurring in former frame, enArea is contiguous range coefficient,Represent STLiTested
Threshold set used when measuring, α represents the coefficient following the tracks of threshold set,Represent STLiLife cycle, MAXLN be STL unit
The maximum lifetime of element, NMSTLNumber of elements for STL, executes following steps:
2-1. is in image srcImgIn, if can allow?'s
Whole characteristic matching successes under constraint;If it can, willIn feature in all srcImg in image update, jump
To 2-2, if it is not, skipping to 2-3;
2-2. uses the hit results in 2-1 to update STLiIn srcImgMake simultaneouslyPlus one, directly
ArriveReach MAXLN, skip to 2-4;
2-3.Subtract one, whenDuring equal to zero, by STLiDelete and by NM from STLSTLSubtract one, skip to 2-4;
2-4.i adds one, if i is more than NMSTLWhen, then terminate;Otherwise, skip to 2-1;
Step 3:Labelling tracing area:
Using the information of the tenacious tracking queue each element being updated in previous step, followed the tracks of in image srcImg
The labelling of target area;
Step 4:The different images that vehicle detection module receives to it carry out heavy duty, then carry out detecting by vehicle detection module
To testing result;
Step 5:Using Pre-tracking queuing method selective mechanisms result:
Obtain vehicle detection result from previous step, screened using Pre-tracking queue, thus that is continuously hit is steady
Surely follow the tracks of target, finally tenacious tracking target is added to tenacious tracking queue as new element, then skip to step 1, until
Photographic head no longer inputs new image;Specifically:
If DRiFor i-th element in detection result sets DR, NMDRFor the quantity of element in detection result sets DR, PTL represents pre-
Tracking queue, PTLjJ-th element for PTL, NMPTLFor the quantity of element in PTL,WithPoint
Wei not DRiFeature set and PTLjFeature set,For PTLjContinuous hit-count,For PTLjIt is whether
The labelling of continuous hit, MAXHTM is the continuous hit-count that detection target can become tenacious tracking target, execution with
Lower step:
5-1. willWithCarry out the coupling of feature, if can not all the match is successful skips to 5-2, such as
Fruit can all the match is successful then skips to 5-3;
J plus one by 5-2., if j is more than NMPTL, then skip to 5-4, otherwise, skip to 5-1;
5-3. willPlus one,It is entered as true, ifEqual to MAXHTM, then by PTLjAs
New element adds STL and deletes from PTL, NMPTLSubtract one, skip to 5-4;
5-4.i adds one, and j is entered as 1, if i is more than NMDRWhen, skip to 5-5;Otherwise, skip to 5-1;
5-5. makes j ∈ [1, NMPTL], for any j, ifValue be not true, then by PTLjFrom Pre-tracking queue
Middle deletion.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410105323.7A CN103886292B (en) | 2014-03-20 | 2014-03-20 | Night vehicle target stable tracking method based on machine vision |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410105323.7A CN103886292B (en) | 2014-03-20 | 2014-03-20 | Night vehicle target stable tracking method based on machine vision |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103886292A CN103886292A (en) | 2014-06-25 |
CN103886292B true CN103886292B (en) | 2017-02-08 |
Family
ID=50955175
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410105323.7A Active CN103886292B (en) | 2014-03-20 | 2014-03-20 | Night vehicle target stable tracking method based on machine vision |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103886292B (en) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105261035B (en) * | 2015-09-15 | 2018-05-11 | 杭州中威电子股份有限公司 | A kind of highway motion target tracking method and device |
CN106530818A (en) * | 2016-12-30 | 2017-03-22 | 北京航空航天大学 | Intelligent parking lot management system based on video processing technology |
CN109521419B (en) * | 2017-09-20 | 2021-07-20 | 比亚迪股份有限公司 | Target tracking method and device based on vehicle radar |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101382997A (en) * | 2008-06-13 | 2009-03-11 | 青岛海信电子产业控股股份有限公司 | Vehicle detecting and tracking method and device at night |
CN101739827A (en) * | 2009-11-24 | 2010-06-16 | 北京中星微电子有限公司 | Vehicle detecting and tracking method and device |
CN102044151A (en) * | 2010-10-14 | 2011-05-04 | 吉林大学 | Night vehicle video detection method based on illumination visibility identification |
CN102385803A (en) * | 2011-10-28 | 2012-03-21 | 南京邮电大学 | All-weather urban vehicle tracking and counting method based on video monitoring |
CN102426783A (en) * | 2011-08-30 | 2012-04-25 | 同济大学 | Low flow road traffic incident detection method based on vehicle tracking |
CN102567705A (en) * | 2010-12-23 | 2012-07-11 | 北京邮电大学 | Method for detecting and tracking night running vehicle |
CN102831426A (en) * | 2012-08-30 | 2012-12-19 | 杭州电子科技大学 | Road environment self-adaptive straight-lane detection method |
CN103150898A (en) * | 2013-01-25 | 2013-06-12 | 大唐移动通信设备有限公司 | Method and device for detection of vehicle at night and method and device for tracking of vehicle at night |
-
2014
- 2014-03-20 CN CN201410105323.7A patent/CN103886292B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101382997A (en) * | 2008-06-13 | 2009-03-11 | 青岛海信电子产业控股股份有限公司 | Vehicle detecting and tracking method and device at night |
CN101739827A (en) * | 2009-11-24 | 2010-06-16 | 北京中星微电子有限公司 | Vehicle detecting and tracking method and device |
CN102044151A (en) * | 2010-10-14 | 2011-05-04 | 吉林大学 | Night vehicle video detection method based on illumination visibility identification |
CN102567705A (en) * | 2010-12-23 | 2012-07-11 | 北京邮电大学 | Method for detecting and tracking night running vehicle |
CN102426783A (en) * | 2011-08-30 | 2012-04-25 | 同济大学 | Low flow road traffic incident detection method based on vehicle tracking |
CN102385803A (en) * | 2011-10-28 | 2012-03-21 | 南京邮电大学 | All-weather urban vehicle tracking and counting method based on video monitoring |
CN102831426A (en) * | 2012-08-30 | 2012-12-19 | 杭州电子科技大学 | Road environment self-adaptive straight-lane detection method |
CN103150898A (en) * | 2013-01-25 | 2013-06-12 | 大唐移动通信设备有限公司 | Method and device for detection of vehicle at night and method and device for tracking of vehicle at night |
Non-Patent Citations (1)
Title |
---|
基于单目视觉的夜间车辆和车距检测;周俊杰;《中国优秀硕士学位论文全文数据库 信息科技辑》;20080115(第01期);I140-251 * |
Also Published As
Publication number | Publication date |
---|---|
CN103886292A (en) | 2014-06-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US20200353932A1 (en) | Traffic light detection method and apparatus, intelligent driving method and apparatus, vehicle, and electronic device | |
CN108230254B (en) | Automatic detection method for high-speed traffic full lane line capable of self-adapting scene switching | |
US11688174B2 (en) | System and method for determining vehicle data set familiarity | |
CN104537841B (en) | Unlicensed vehicle violation detection method and detection system thereof | |
CN110532896A (en) | A kind of road vehicle detection method merged based on trackside millimetre-wave radar and machine vision | |
CN109190444B (en) | Method for realizing video-based toll lane vehicle feature recognition system | |
CN105513349B (en) | Mountainous area highway vehicular events detection method based on double-visual angle study | |
CN103903019A (en) | Automatic generating method for multi-lane vehicle track space-time diagram | |
WO2021026120A1 (en) | Using captured video data to identify active turn signals on a vehicle | |
CN111667512A (en) | Multi-target vehicle track prediction method based on improved Kalman filtering | |
US11900690B2 (en) | Apparatus, method, and computer program for identifying state of signal light, and controller | |
CN110422173B (en) | Driving environment identification method | |
CN103886292B (en) | Night vehicle target stable tracking method based on machine vision | |
CN103310206B (en) | A kind of vehicle using motor detection method based on many features and multiframe information fusion | |
CN104951758A (en) | Vehicle-mounted method and vehicle-mounted system for detecting and tracking pedestrians based on vision under urban environment | |
CN107506753B (en) | Multi-vehicle tracking method for dynamic video monitoring | |
CN110555423B (en) | Multi-dimensional motion camera-based traffic parameter extraction method for aerial video | |
Wei et al. | Street object detection/tracking for AI city traffic analysis | |
CN117275216A (en) | Multifunctional unmanned aerial vehicle expressway inspection system | |
Zhao et al. | An Improved Method for Infrared Vehicle and Pedestrian Detection Based on YOLOv5s | |
Baiat et al. | Smart City Traffic Monitoring: YOLOv7 Transfer Learning Approach for Real-Time Vehicle Detection | |
CN112116634B (en) | Multi-target tracking method of semi-online machine | |
Mijic et al. | Autonomous driving solution based on traffic sign detection | |
Du et al. | A Vision-based Taillight Intention Detection Method for Intelligent and Connected Vehicles | |
Kanjee et al. | A three-step vehicle detection framework for range estimation using a single camera |
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 |