CN105488484B - A kind of track of vehicle extracting method based on unmanned plane image - Google Patents
A kind of track of vehicle extracting method based on unmanned plane image Download PDFInfo
- Publication number
- CN105488484B CN105488484B CN201510889262.2A CN201510889262A CN105488484B CN 105488484 B CN105488484 B CN 105488484B CN 201510889262 A CN201510889262 A CN 201510889262A CN 105488484 B CN105488484 B CN 105488484B
- Authority
- CN
- China
- Prior art keywords
- vehicle
- frame
- target vehicle
- image
- track
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/08—Detecting or categorising vehicles
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of track of vehicle extracting methods based on unmanned plane image, including step 1：Vehicle is detected based on Haar classifier；Step 2：Based on the doubtful candidate vehicle of position prediction screening；Step 3：Target vehicle is determined based on 4 channel color histogram similarities；Step 4：Track of vehicle extracts；The more color histogram similarities of innovative joint of the invention carry out target vehicle screening, have very high discrimination to same vehicle and other vehicles；The track of vehicle extracting method can be restored to track automatically in the case where vehicle tracking interrupts.
Description
Technical field
The invention belongs to technical field of image processing, it is related to a kind of track of vehicle based on unmanned plane low latitude Aerial Images and mentions
Method is taken, traffic monitoring field is suitable for.
Background technique
Unmanned plane has very big application prospect in traffic monitoring field, for example carries out under emergency event using unmanned plane
Rapid transit Situation Awareness.Existing traffic monitoring mode is mainly by the camera, micro- for being installed on fixed point (such as electric pole)
Wave radar equal section detection mode can not obtain the traffic state data of macroscopic view, and the place for not installing monitoring device can not be obtained
Take traffic behavior.Unmanned plane itself is a mobile data acquisition platform, relative to fixed traffic monitoring mode, has machine
The advantage that dynamic property is strong, the visual field is wide, flight is not limited by ground, therefore can realize that macro-traffic status data extracts, for not pacifying
Traffic monitoring also may be implemented in the section for filling fixed traffic monitoring apparatus.
Summary of the invention
The purpose of the present invention is to solve the above problems, propose a kind of track of vehicle based on unmanned plane image
Extracting method is primarily based on Haar feature cascade classifier and carries out vehicle detection, is then based on position prediction and multichannel color
Histogram similarity realizes vehicle tracking, realizes that track of vehicle extracts.
The present invention studies access point with completely new, proposes for unmanned plane low latitude Aerial Images a kind of based on unmanned plane
The track of vehicle extracting method of image, is realized by following step：
Step 1：Vehicle is detected based on Haar classifier；
For unmanned plane low latitude aerial image sequence, vehicle detection, Haar classifier are carried out using based on Haar classifier
It needs to be trained based on sample database, sample database is the image data set comprising positive negative sample, is needed by hand from including just
It is chosen in the image of negative sample.Vehicle detection is carried out based on trained Haar classifier, obtains vehicle in Aerial Images
Pixel coordinate, size (length and width) and frame number, then obtain the channel R, the channel G, channel B and the gray scale of the vehicle RGB color space
The color histogram in Fig. 4 channel, the vehicle for carrying out different interframe in subsequent step are associated with.
Step 2：Based on the doubtful candidate vehicle of position prediction screening；
Taking certain vehicle being detected in former frame is target vehicle to be tracked, and is delimited centered on the position of the vehicle
One rectangle estimation range (default size be target vehicle twice), its center of vehicle being detected in present frame fall into this
Rectangle estimation range all as the suspected target vehicle of target vehicle；
Step 3：Target vehicle is determined based on 4 channel color histogram similarities；
The doubtful candidate vehicle screened in upper step generally has 1~2, to confirm target vehicle, passes through 4 channel colors
Histogram similarity is screened.Successively calculate the color histogram similarity value of target vehicle and doubtful candidate vehicle, and root
It is ranked up according to size, selectes maximum similarity, if the threshold value of the big Mr. Yu's setting of the value, then it is assumed that with the similitude
Being worth corresponding doubtful candidate vehicle-to-target vehicle is same vehicle, by frame number, the coordinate, ruler in doubtful candidate vehicle present frame
Very little size, color histogram information pass to target vehicle, complete target vehicle information update, that is, realize the tracking of vehicle.
If the threshold value of the maximum small Mr. Yu's setting of similarity, then it is assumed that do not find target vehicle, i.e. target vehicle in this frame image
Tracking failure.The vehicle that failure is tracked in present frame will re-start search in the next frame, and difference is, in order to search
The vehicle of failure is tracked, the rectangle estimation range size centered on the position of the vehicle in a previous frame will do it amplification (default ruler
Very little four times for target vehicle), to increase the probability for searching tracking failure vehicle.When the vehicle of tracking failure is subsequent continuous
When can not all be tracked in 3 frame images, algorithm can abandon processing to the vehicle, no longer track, the advantage of doing so is that can be with
Shield some misjudged break as the non-vehicle object of vehicle.
Step 4：Track of vehicle extracts；
When tracked vehicle disappears from image, that is, terminate to track, and position of the vehicle in different frame, frame number is defeated
It saves out, that is, completes track of vehicle extraction.
The advantage of the invention is that：
(1) there is unmanned plane image the wide visual field, mobility strong, flight not to be influenced by landform, relative to based on solid
The track of vehicle extracting method for determining camera, the track of vehicle extracting method based on unmanned plane image have vehicle will not be by
It blocks, there is no the advantages of perspective distortion.Position predicting method of the present invention can substantially reduce the doubtful vehicle of target
Region of search greatly reduces the quantity of the doubtful vehicle of target, reduces the differentiation difficulty of target vehicle Yu doubtful vehicle, improves
Vehicle tracking precision；
(2) the more color histogram similarities of the innovative joint of the present invention carry out target vehicle screenings, to same vehicle with
Other vehicles have very high discrimination；The track of vehicle extracting method, can be certainly in the case where vehicle tracking interrupts
It is dynamic to restore tracking；
(3) track of vehicle extraction algorithm of the present invention can track simultaneously and detect dozens to hundreds of vehicle rail
Mark has breakthrough innovation.
Detailed description of the invention
Fig. 1 is the HAAR feature that vehicle detection is used in the present invention；
Fig. 2 is to carry out vehicle detection based on HAAR feature；
Fig. 3 is the sample image for training Haar classifier；
Fig. 4 is position prediction figure；
Fig. 5 is 4 channel color histogram of vehicle；
Fig. 6 is flow chart of the method for the present invention.
Specific embodiment
Below in conjunction with drawings and examples, the present invention is described in further detail.
The present invention provides a kind of track of vehicle extracting method based on unmanned plane image, and the method is primarily based on
Haar classifier carries out processing detection vehicle to image sequence, is then based on position prediction and multichannel color histogram similarity
It realizes vehicle tracking, then realizes that track of vehicle extracts.Here it is extracted with the track of vehicle of a vehicle to introduce and be based on figure of taking photo by plane
The highway vehicle trajectory extraction algorithm of picture, the trajectory extraction process of single vehicle is as shown in fig. 6, specific processing step is as follows：
Step 1：Vehicle is detected based on Haar classifier；
For unmanned plane low latitude aerial image sequence, vehicle detection, Haar used are carried out using based on Haar classifier
Feature be it is shown in FIG. 1, the vehicle detection effect based on Haar classifier is as shown in Figure 2.The Haar classifier is detecting
Firstly the need of being trained based on positive negative sample before vehicle, training method used is adaboost algorithm, and wherein positive sample is such as
Shown in Fig. 3 left figure, negative sample is as shown in Fig. 3 right figure.Vehicle is detected using Haar classifier, chooses either objective therein here
Vehicle Oi, obtain OiPixel coordinate P in this frame imagei(xi,yi), wide height (wi,hi), and calculate the face in four channels of vehicle
Color Histogram.
Step 2：Based on the doubtful candidate vehicle of position prediction screening；
As shown in the position prediction figure of Fig. 4, to extract target vehicle OiTrack, it need to be tracked.Based on Haar
Detection of classifier next frame image is based on Pi(xi,yi), wide height (wi,hi) setting rectangle estimation range R (such as Fig. 4), wherein the
The definition of R is in t+1 frame：
R:{(xt+1,yt+1)|xt+1∈[xi±αwi]∩[1,ImgW],yt+1∈[yi±αhi]∩[1,ImgH]} (1)
Wherein：Wherein (xt+1,yt+1) it is any point in the R of rectangle estimation range, which meets defined in formula (1)
Condition.ImgW and ImgH is the width and height of unmanned plane image, and α is a scale factor, default value 1.
To the vehicle that t+1 frame detects, vehicle center falls into all vehicles of rectangle estimation range R all as doubtful
Like target vehicle, show there are 2 suspected target vehicles in t+1 frame in Fig. 4.
Step 3：Target vehicle O is determined based on the color histogram similarity in 4 channelsi；
To two suspected target vehicles in step 2, need to judge which vehicle is target vehicle O in this two carsi, or
Two cars are not target vehicle Oi.Here suspected target vehicle-to-target vehicle is calculated based on color histogram match method
Similitude.The relative frequency h that the pixel that intensity value is i in certain single channel image P occursi(P) calculation method is shown below：
Wherein：si(P) be the pixel that intensity value is i in single channel image P number, n be pixel value the total rank of gray scale
Number.The relative frequency h occurred based on every kind of intensity value of image P1(P)~hn(P), the histogram H (P) of single channel image P can be obtained
=[h1(P),h2(P),…,hn(P)].It is identical as the calculation method of histogram H (P) of single channel image P, successively it is calculated
Color image R, G, B, the histogram H (R) in tetra- channels gray scale Gray, H (G), H (B), H (Gray).
Wherein the color histogram in 4 channels of different vehicle is as shown in Figure 5.The corresponding single channel image i's of two cars
The similarity C of color histogramiThe calculation method of (P, P ') is shown below：
Wherein：N is the total order of gray scale of pixel value；I ∈ { R, G, B, Gray } represents tetra- R, G, B, gray scale Gray single-passes
One of road image；P ' and P respectively indicate single channel image corresponding to the channel i of two cars (color image)；
For the number for the pixel that color intensity value in image P is j,For the flat of the corresponding pixel of intensity values all in image P
Equal number, whereinWherein, CiFor the value of (P, P ') between 0 and 1,0 represents two single channel figures
As histogram it is completely dissimilar, 1 represent it is identical.
Four channel R, G of two cars (color image), B, face corresponding to gray scale Gray is successively calculated according to formula (3)
Color Histogram similar value CR(P,P′)、CG(P,P′)、CB(P, P ') and CGray(P, P '), then 4 channels of two cars is composite coloured
Histogram similar value S (P, P ') can be obtained, and be shown below：
S (P, P ')=∏i∈{R,G,B,Gray}Ci(P,P′) (4)
The doubtful candidate vehicles of several filtered out in step 2 based on position prediction, are calculated separately and target carriage by formula (4)
OiComposite coloured histogram similar value S (P, P '), can obtain multiple similar values here, choose maximum value, if should
Value is greater than a preset threshold value, then it is assumed that doubtful candidate's vehicle-to-target vehicle O corresponding to the valueiIt is same vehicle, so
Algorithm can record as shown in Figure 4 and update target vehicle position in the current frame, wide high, time series (frame number) and 4 afterwards
The color histogram in channel.Target vehicle is completed in the tracking of present frame, then proceedes to repeat step 1~step 3, realization pair
Vehicle is continuously tracked.If the maximum value in obtained multiple composite coloured histogram similar values is less than preset threshold value, recognize
Not find target vehicle O in this frame imagei, judge the coordinate position of the last moment of this vehicle whether in image at this time
Edge, if be in image edge, that is, think that the vehicle has left field of view, i.e., tracking complete, enter step 4 progress at this time
Trajectory extraction.If the coordinate position of the last moment of this vehicle thinks target vehicle O not at the edge of imageiIn the frame
Tracking failure enters next frame at this time and continues vehicle detection, next proceeds through the doubtful candidate vehicle of position prediction screening,
Amplification (default size be target vehicle four times) will do it for vehicle its rectangle estimation range size of tracking failure, with increasing
The probability of tracking failure vehicle is searched greatly.When the vehicle of tracking failure is found again, rectangle estimation range can be reverted to
Default size.When the vehicle of tracking failure can not be all tracked in subsequent continuous 3 frame image, algorithm can throw the vehicle
Abandoning processing, no longer tracks, the advantage of doing so is that some misjudged break as the non-vehicle object of vehicle can be shielded.
Step 4：Track of vehicle extracts；
When a tracked vehicle is lost in certain frame image, judge that the coordinate position of the last moment of this vehicle is at this time
The no edge in image thinks that the vehicle has left field of view if being in the edge of image, i.e. tracking is completed, and terminates at this time
Tracking, the algorithm can export the coordinate of the vehicle recorded, time serial message, that is, realize track of vehicle extraction.
Claims (3)
1. a kind of track of vehicle extracting method based on unmanned plane image, including following steps：
Step 1：Vehicle is detected based on Haar classifier；
For unmanned plane image sequence, vehicle detection is carried out based on trained Haar classifier, vehicle is obtained and is taking photo by plane
Then pixel coordinate, size and frame number in image obtain the channel R, the channel G, channel B and the gray scale of the vehicle RGB color space
The color histogram in Fig. 4 channel；
Step 2：Based on the doubtful candidate vehicle of position prediction screening；
Taking certain vehicle being detected in former frame is target vehicle to be tracked, and one delimited centered on the position of the vehicle
Rectangle estimation range, its center of vehicle being detected in present frame fall into the rectangle estimation range all as target vehicle
Suspected target vehicle；
Step 3：Target vehicle is determined based on 4 channel color histogram similarities；
For present frame, suspected target vehicle is screened by 4 channel color histograms, successively calculate target vehicle and is doubted
It like the color histogram similarity value of candidate vehicle, and is ranked up according to size, maximum similarity is selected, if most
Big similarity is greater than the threshold value of setting, then it is assumed that doubtful candidate vehicle-to-target vehicle corresponding with the similarity is same
Frame number, coordinate, size, the color histogram information of doubtful candidate vehicle in the current frame are passed to target by one vehicle
Vehicle, carry out target vehicle information update, think that target vehicle tracks success in the frame at this time, fetch it is lower enter next frame,
Continue target vehicle tracking；
If maximum similarity is less than the threshold value of setting, then it is assumed that do not find target vehicle in this frame image, judge target
The coordinate position of the last moment of vehicle whether at the edge of image, if be in image edge, that is, think target vehicle from
Field of view is opened, i.e. tracking is completed, and 4 carry out trajectory extractions are entered step；If the coordinate position of the last moment of target vehicle is not
At the edge of image, that is, think that target vehicle tracks failure in the frame；The vehicle that failure is tracked in present frame will be in next frame
In re-start search, rectangle estimation range ruler when scanning for, in next frame centered on vehicle position in a previous frame
It is very little to amplify, if target vehicle is found again, rectangle estimation range revert to amplification before size, into next frame into
The tracking of row target vehicle；If target vehicle tracking failure, rectangle estimation range size constancy re-start search in the next frame
Rope abandons processing to the vehicle, no longer when the vehicle of tracking failure can not be all tracked in subsequent continuous 3 frame image
Tracking；
It all searches for and finishes when frame each in image sequence, terminate tracking, enter step 4；
Step 4：Track of vehicle extracts；
According to position of the obtained vehicle in different frame, frame number, completes track of vehicle and extract.
2. a kind of track of vehicle extracting method based on unmanned plane image according to claim 1, the step 2
In, twice having a size of target vehicle size of rectangle estimation range.
3. a kind of track of vehicle extracting method based on unmanned plane image according to claim 1, the step 3
In, the rectangle estimation range of amplification is having a size of the four of target vehicle size times.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510889262.2A CN105488484B (en) | 2015-12-07 | 2015-12-07 | A kind of track of vehicle extracting method based on unmanned plane image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510889262.2A CN105488484B (en) | 2015-12-07 | 2015-12-07 | A kind of track of vehicle extracting method based on unmanned plane image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105488484A CN105488484A (en) | 2016-04-13 |
| CN105488484B true CN105488484B (en) | 2018-11-27 |
Family
ID=55675456
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510889262.2A Active CN105488484B (en) | 2015-12-07 | 2015-12-07 | A kind of track of vehicle extracting method based on unmanned plane image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105488484B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106874854B (en) * | 2017-01-19 | 2019-12-31 | 西安电子科技大学 | Unmanned aerial vehicle tracking method based on embedded platform |
| CN108009494A (en) * | 2017-11-30 | 2018-05-08 | 中山大学 | A kind of intersection wireless vehicle tracking based on unmanned plane |
| CN108960190B (en) * | 2018-07-23 | 2021-11-30 | 西安电子科技大学 | SAR video target detection method based on FCN image sequence model |
| CN109754441A (en) * | 2019-01-10 | 2019-05-14 | 海南大学 | Ship tracking based on position prediction and color similarity |
| CN110221611B (en) * | 2019-06-11 | 2020-09-04 | 北京三快在线科技有限公司 | Trajectory tracking control method and device and unmanned vehicle |
| CN114495545A (en) * | 2022-01-28 | 2022-05-13 | 常州海蓝利科物联网技术有限公司 | Vehicle control system and method |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2003344073A (en) * | 2002-05-22 | 2003-12-03 | Denso Corp | Navigation apparatus and program |
| CN102358287A (en) * | 2011-09-05 | 2012-02-22 | 北京航空航天大学 | Trajectory tracking control method used for automatic driving robot of vehicle |
| CN103116986A (en) * | 2013-01-21 | 2013-05-22 | 信帧电子技术（北京）有限公司 | Vehicle identification method |
| CN103150547A (en) * | 2013-01-21 | 2013-06-12 | 信帧电子技术（北京）有限公司 | Vehicle tracking method and device |
| CN103207988A (en) * | 2013-03-06 | 2013-07-17 | 大唐移动通信设备有限公司 | Method and device for image identification |
| CN103903019A (en) * | 2014-04-11 | 2014-07-02 | 北京工业大学 | Automatic generating method for multi-lane vehicle track space-time diagram |
| CN104951784A (en) * | 2015-06-03 | 2015-09-30 | 杨英仓 | Method of detecting absence and coverage of license plate in real time |
| CN104992145A (en) * | 2015-06-15 | 2015-10-21 | 山东大学 | Moment sampling lane tracking detection method |
-
2015
- 2015-12-07 CN CN201510889262.2A patent/CN105488484B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2003344073A (en) * | 2002-05-22 | 2003-12-03 | Denso Corp | Navigation apparatus and program |
| CN102358287A (en) * | 2011-09-05 | 2012-02-22 | 北京航空航天大学 | Trajectory tracking control method used for automatic driving robot of vehicle |
| CN103116986A (en) * | 2013-01-21 | 2013-05-22 | 信帧电子技术（北京）有限公司 | Vehicle identification method |
| CN103150547A (en) * | 2013-01-21 | 2013-06-12 | 信帧电子技术（北京）有限公司 | Vehicle tracking method and device |
| CN103207988A (en) * | 2013-03-06 | 2013-07-17 | 大唐移动通信设备有限公司 | Method and device for image identification |
| CN103903019A (en) * | 2014-04-11 | 2014-07-02 | 北京工业大学 | Automatic generating method for multi-lane vehicle track space-time diagram |
| CN104951784A (en) * | 2015-06-03 | 2015-09-30 | 杨英仓 | Method of detecting absence and coverage of license plate in real time |
| CN104992145A (en) * | 2015-06-15 | 2015-10-21 | 山东大学 | Moment sampling lane tracking detection method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105488484A (en) | 2016-04-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105488484B (en) | A kind of track of vehicle extracting method based on unmanned plane image | |
| US20170032676A1 (en) | System for detecting pedestrians by fusing color and depth information | |
| US8750567B2 (en) | Road structure detection and tracking | |
| CN103824070B (en) | A kind of rapid pedestrian detection method based on computer vision | |
| Cao et al. | Vehicle detection and motion analysis in low-altitude airborne video under urban environment | |
| CN103871077B (en) | A kind of extraction method of key frame in road vehicles monitoring video | |
| US9626599B2 (en) | Reconfigurable clear path detection system | |
| EP2919189A2 (en) | Pedestrian tracking and counting method and device for near-front top-view monitoring video | |
| KR101395094B1 (en) | Method and system for detecting object in input image | |
| CN106448160B (en) | Target person tracking method combining vehicle running track and monitoring video data | |
| JP5931662B2 (en) | Road condition monitoring apparatus and road condition monitoring method | |
| CN102622884A (en) | Vehicle illegal turning behavior detection method based on tracking | |
| Maria et al. | A drone-based image processing system for car detection in a smart transport infrastructure | |
| CN104463903A (en) | Pedestrian image real-time detection method based on target behavior analysis | |
| Lim et al. | Lane detection and Kalman-based linear-parabolic lane tracking | |
| CN104463232A (en) | Density crowd counting method based on HOG characteristic and color histogram characteristic | |
| CN104036250A (en) | Video pedestrian detecting and tracking method | |
| CN101369312B (en) | Method and equipment for detecting intersection in image | |
| CN105426816A (en) | Method and device of processing face images | |
| CN104392461A (en) | Video tracking method based on texture features | |
| Shine et al. | Fractional data distillation model for anomaly detection in traffic videos | |
| CN110443142B (en) | Deep learning vehicle counting method based on road surface extraction and segmentation | |
| Cao et al. | Visual attention accelerated vehicle detection in low-altitude airborne video of urban environment | |
| CN112347967B (en) | Pedestrian detection method fusing motion information in complex scene | |
| CN109255052A (en) | A kind of three stage vehicle retrieval methods based on multiple features |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |