CN103886609B - Vehicle tracking method based on particle filtering and LBP features - Google Patents
Vehicle tracking method based on particle filtering and LBP features Download PDFInfo
- Publication number
- CN103886609B CN103886609B CN201410133577.XA CN201410133577A CN103886609B CN 103886609 B CN103886609 B CN 103886609B CN 201410133577 A CN201410133577 A CN 201410133577A CN 103886609 B CN103886609 B CN 103886609B
- Authority
- CN
- China
- Prior art keywords
- particle
- lbp
- vehicle
- tracking
- frames
- 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 vehicle tracking method based on particle filtering and LBP textural features. On the basis of a particle filtering algorithm, vehicles are tracked and detected by using an LBP feature histogram, and the influence of light on tracking is reduced. The feature that the LBP feature gray level is not changed is fully utilized, so that the influence of light variation, shadow interference and the like on tracking is weakened, a plurality of vehicles can be effectively tracked, vehicle traveling information can be extracted, and therefore data support is provided for obtainment of the basic information such as the number of the vehicles, the positions of the vehicles, the models of the vehicles, the speeds of the vehicles and the like in the next step, and for vehicle behavior explanation and traffic jam analysis.
Description
Technical field
The present invention relates to it is a kind of based on particle filter and the wireless vehicle tracking of LBP features, belong to computer vision field.
Background technology
For traffic administration personnel, the video image of traffic intersection is most intuitively transport information, while being also maximum
Source of traffic information.Containing abundant transport information in video.With Digital Image Processing, compunication, pattern-recognition and people
The development of the subjects such as work intelligence, is achieved based on the traffic data collection technology of video and is developed rapidly.Traffic number based on video
According to acquisition technique with video image to analyze object, processed and analyzed by the image to setting regions, exactly from regarding
Moving vehicle is detected in frequency image, vehicle tracking and identification are carried out on this basis, finally obtain vehicle fleet size, position, car
The essential informations such as type, speed, for vehicle behavior explanation, traffic congestion analysis data supporting is provided.Traffic data based on video
Collection represents traffic information collection with the developing direction for processing.Therefore moving vehicle detection and tracking are intelligent monitor systems one
Basis and the work of key.
But, the impact and the complexity of traffic scene due to natural environment, the moving object detection in traffic scene and
Tracking result often by weather, light, shooting angle, target sizes, shade, block and affected larger with target speed.
Therefore, moving object detection and follow-up study in traffic video monitoring is the important research field of intelligent transportation system, is had
Major and immediate significance.Vehicle tracking has become computer vision field one important as one of important step
Problem.
At present, the method for vehicle tracking mainly has based on the tracking of auto model, based on the tracking of vehicle region, based on car
The tracking in region, the tracking based on vehicle's contour.Based on the tracking of auto model, core is car known to accurate extraction
3-D models, and cause altimetric image to be checked match with model library.The advantage of the method is that the degree of accuracy is high, and shortcoming is to mould
Type is excessively relied on, and because introducing model library, causes to produce very big amount of calculation, thus the method use it is less.Based on vehicle
The tracking in region, is to represent vehicle in the form of block, then join domain is merged or is split.The method exists
Effect is fine in the case of vehicle rareness, but when wagon flow is intensive, for there is inaccuracy in the combination and segmentation in region.Based on vehicle
The tracking of profile, is first to extract vehicle's contour, and is updated in follow-up each two field picture.The method is substantially based on
The improvement of region method, reduces amount of calculation, but does not solve the problems, such as effect on driving birds is not good under shade and congestion situation.Based on vehicle
Feature tracking, be as minimum tracking cell, and by these combinations of features representing vehicle using the feature of vehicle.Should
Method has the prominent advantages that and efficiently solves occlusion issue, light variation issue, and is transplanted to the efficiency after embedded platform
Problem, but effective selected characteristic or merge multiple features for how, all also many problems need research.
The content of the invention
The purpose of the present invention is to overcome the deficiencies in the prior art, there is provided a kind of based on particle filter and LBP textural characteristics
Wireless vehicle tracking, solves because light changes the problems such as reducing tracking accuracy rate.
According to the technical scheme that the present invention is provided, the wireless vehicle tracking based on particle filter and LBP features includes
Following steps:
Step 1, read the first two field picture from the video image for collecting and be set to pre frames, manually demarcate in the picture
Go out to track the rectangular area at vehicle place, in this, as the template that next frame is compared;
Step 2, will pre two field pictures back up after be converted to gray-scale map, then calculate demarcate region LBP feature histograms;
Step 3, the next two field picture of reading are simultaneously set to p frames, will be converted to gray-scale map after the backup of p two field pictures;
Step 4, step 3 obtain gray-scale map image in by Gaussian Profile arrange np particle carry out particle filter;Grain
Son filtering is referred to be carried out approximately, with sample by finding one group of random sample propagated in state space to probability density function
Average replaces integral operation, and so as to obtain the process of state minimum variance estimate, these samples are referred to as particle;
The LBP feature histograms of np particle, make each neighborhood picture in template and p frames in step 5, calculating pre two field pictures
Vegetarian refreshments is poor with center pixel, and more than 01 is put, and sets to 0 less than 0, and 0/1 sequence is constructed counterclockwise, and it is right finally to calculate
The decimal number answered, is the LBP texture eigenvalues of the pixel, and to each pixel in demarcation region LBP characteristic values are sought, and
Counted, that is, obtained the LBP feature histograms of image;
Step 6, update after systematic observation particle weights so that its weight is matched with physical possibility, i.e. next step
The possibility that the particle occurs is corresponding according to corresponding proportion with weight, and implementation method is to use the LBP of template in image pre frames special
Levy histogram carries out variance calculating with the LBP feature histograms of particle in image p frames, and is standardized, and each grain is obtained respectively
Weight w of son;
Step 7, present frame is set to pre frames, the maximum particle of weight w value is target position, and set should
Particle is new template, for comparison next time;
Step 8, carry out re-sampling operations, it is therefore an objective to replicated on the particle of previous step, so as to respond a fixed number is produced
The particle of amount;The quantity of duplication is calculated according to np*w;Then repeating step 2~7 carries out the tracking of next frame.If w<
0.1 then no longer replicates the particle.
In step 1, when there is multiple stage tracking vehicle, manually calibrate in the picture and track the rectangle region that vehicle is located per platform
Domain, as the multiple template that next frame is compared.
It is an advantage of the invention that:This patent is tracked using particle filter algorithm, it is possible to achieve to the demand for tracking, and
By abandoning traditional hsv color model, selective extraction LBP features, so as to weaken light change, shadow interference etc. to tracking
Affect.This patent can effectively track multiple stage vehicle, extract vehicle traveling information, so as to obtain vehicle fleet size, position for next step
Put, the essential information such as vehicle, speed, provide data supporting for vehicle behavior explanation, traffic congestion analysis.
Description of the drawings
Fig. 1 is LBP feature calculation method schematic diagrams.
Fig. 2 is particle filter implementation procedure schematic diagram.
Fig. 3 is the implementation steps flow chart of the present invention.
Specific embodiment
With reference to the accompanying drawings and examples the invention will be further described.
This patent be directed to it is a kind of based on particle filter and the wireless vehicle tracking of LBP features, by using LBP features
Histogram, tracing detection reduces impact of the light to tracking to vehicle, and idiographic flow is as shown in Figure 3.
Step 1, setting number of particles np(np<100), the first two field picture is read from the video image for collecting and is set to
Pre frames, manually calibrate in the picture tracking vehicle region, and the region is rectangle, can be just about to the vehicle for tracking
In being enclosed in, in this, as the template that next frame is compared.When having multiple stage tracking vehicle, then calibrate respectively and track vehicle institute per platform
Rectangular area, as next frame compare multiple template.
Step 2, will pre two field pictures back up after be converted to gray-scale map, then calculate demarcate region LBP feature histograms.
As shown in figure 1, there are 8 pixels to participate in computing, in the case that radius is 2 pixels, by Fig. 1 left figure template thresholdings, each neighbour is made
Domain pixel is made comparisons with center pixel, more than then putting 1, less than then setting to 0, obtains scheming in Fig. 1, and 0/1 is constructed counterclockwise
Sequence(10100101), corresponding decimal number 165 is finally calculated, the LBP texture eigenvalues of the pixel are exactly 165.
Step 3, the next two field picture of reading are simultaneously set to p frames, by backup image, the gray-scale map for then converting the image into.
Step 4, in gray-scale map image by Gaussian Profile arrange np particle carry out particle filter;It is logical that particle filter refers to
Cross one group of searching random sample of propagation in state space is carried out approximately to probability density function, replaces integrating with sample average
Computing, so as to obtain the process of state minimum variance estimate, these samples are referred to as " particle " by image.As shown in Fig. 2 we
12 examples are chosen, and in the first secondary tracking particle is arranged in around the template obtained in step one according to Gaussian Profile.
Then the similarity of particle and template is calculated, and the weight of all particles is normalized, in tracking next time, each position
The number of particles of arrangement is the product of total number of particles and weight.Later particle placement process is same.
The LBP feature histograms of np particle in template and p frames in step 5, calculating pre two field pictures, method is with step 2
Computational Methods are identical.
Step 6, update after systematic observation particle weights so that its weight is matched with physical possibility, i.e. next step
The possibility that the particle occurs is corresponding according to corresponding proportion with weight, and implementation method is to use the LBP of template in image pre frames special
Levy histogram carries out variance calculating with the LBP feature histograms of particle in image p frames, and is normalized(I.e. all particle power
Weight sum is 1), weight w of each particle is obtained respectively.
Step 7, present frame is set to pre frames, the maximum particle of weight w value is target position, and set should
Particle is new template, for comparison next time.
Step 8, carry out re-sampling operations, it is therefore an objective to replicated on the particle of previous step, so as to respond a fixed number is produced
The particle of amount.The quantity of duplication is calculated according to np*w, if w<0.1 then no longer replicates the particle;Then step 2 is repeated,
3,4,5,6,7 tracking for carrying out next frame.
Claims (1)
1. based on particle filter and the wireless vehicle tracking of LBP features, it is characterized in that, comprise the following steps:
Step 1, read the first two field picture from the video image for collecting and be set to pre frames, manually calibrate in the picture with
The rectangular area that track vehicle is located, in this, as the template that next frame is compared;
Step 2, will pre two field pictures back up after be converted to gray-scale map, then calculate demarcate region LBP feature histograms;
Step 3, the next two field picture of reading are simultaneously set to p frames, will be converted to gray-scale map after the backup of p two field pictures;
Step 4, step 3 obtain gray-scale map image in by Gaussian Profile arrange np particle carry out particle filter;Particle is filtered
Ripple is referred to be carried out approximately, with sample average by finding one group of random sample propagated in state space to probability density function
Replace integral operation, so as to obtain the process of state minimum variance estimate, these samples are referred to as particle;
The LBP feature histograms of np particle, make each neighborhood territory pixel point in template and p frames in step 5, calculating pre two field pictures
It is poor with center pixel, 1 is put more than 0, set to 0 less than 0,0/1 sequence is constructed counterclockwise, finally calculate corresponding
Decimal number, is the LBP texture eigenvalues of the pixel, and to each pixel in demarcation region LBP characteristic values are sought, and is carried out
Statistics, that is, obtain the LBP feature histograms of image;
Step 6, update after systematic observation particle weights so that its weight is matched with physical possibility, i.e. the next step grain
The possibility that son occurs is corresponding according to corresponding proportion with weight, and implementation method is to use the LBP features of template in image pre frames straight
Side's figure carries out variance calculating with the LBP feature histograms of particle in image p frames, and is standardized, and each particle is obtained respectively
Weight w;
Step 7, present frame is set to pre frames, the maximum particle of weight w value is target position, and sets the particle
For new template, for comparison next time;
Step 8, carry out re-sampling operations, it is therefore an objective to replicated on the particle of previous step, so as to respond produce it is a number of
Particle;The quantity of duplication is calculated according to np*w;Then repeating step 2~7 carries out the tracking of next frame;
If w in step 8<0.1 then no longer replicates the particle;
In the step 1, when there is multiple stage tracking vehicle, manually calibrate in the picture and track the rectangle region that vehicle is located per platform
Domain, as the multiple template that next frame is compared.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410133577.XA CN103886609B (en) | 2014-04-03 | 2014-04-03 | Vehicle tracking method based on particle filtering and LBP features |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410133577.XA CN103886609B (en) | 2014-04-03 | 2014-04-03 | Vehicle tracking method based on particle filtering and LBP features |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103886609A CN103886609A (en) | 2014-06-25 |
CN103886609B true CN103886609B (en) | 2017-04-12 |
Family
ID=50955482
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410133577.XA Active CN103886609B (en) | 2014-04-03 | 2014-04-03 | Vehicle tracking method based on particle filtering and LBP features |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103886609B (en) |
Families Citing this family (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104200226B (en) * | 2014-09-01 | 2017-08-25 | 西安电子科技大学 | Particle filter method for tracking target based on machine learning |
CN104392461B (en) * | 2014-12-17 | 2017-07-11 | 中山大学 | A kind of video tracing method based on textural characteristics |
CN105741324A (en) * | 2016-03-11 | 2016-07-06 | 江苏物联网研究发展中心 | Moving object detection identification and tracking method on moving platform |
CN108776972B (en) * | 2018-05-04 | 2020-06-12 | 北京邮电大学 | Object tracking method and device |
CN109540149A (en) * | 2018-12-19 | 2019-03-29 | 北京交通大学 | A method of real-time tracing being carried out to indoor automobile using VeTrack system |
WO2020211047A1 (en) * | 2019-04-18 | 2020-10-22 | 京东方科技集团股份有限公司 | Traffic information processing device, system and method |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101577005A (en) * | 2009-06-12 | 2009-11-11 | 北京中星微电子有限公司 | Target tracking method and device |
-
2014
- 2014-04-03 CN CN201410133577.XA patent/CN103886609B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101577005A (en) * | 2009-06-12 | 2009-11-11 | 北京中星微电子有限公司 | Target tracking method and device |
Non-Patent Citations (1)
Title |
---|
融合典型纹理特征的粒子滤波目标跟踪方法;吴刚;《计算机工程与应用》;20111231;第47卷(第34期);21-25 * |
Also Published As
Publication number | Publication date |
---|---|
CN103886609A (en) | 2014-06-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109816024B (en) | Real-time vehicle logo detection method based on multi-scale feature fusion and DCNN | |
CN103886609B (en) | Vehicle tracking method based on particle filtering and LBP features | |
Soilán et al. | Traffic sign detection in MLS acquired point clouds for geometric and image-based semantic inventory | |
CN110188807B (en) | Tunnel pedestrian target detection method based on cascading super-resolution network and improved Faster R-CNN | |
Hadi et al. | Vehicle detection and tracking techniques: a concise review | |
CN107766821B (en) | Method and system for detecting and tracking full-time vehicle in video based on Kalman filtering and deep learning | |
CN108921083B (en) | Illegal mobile vendor identification method based on deep learning target detection | |
CN111444821A (en) | Automatic identification method for urban road signs | |
CN104978567B (en) | Vehicle checking method based on scene classification | |
Derpanis et al. | Classification of traffic video based on a spatiotemporal orientation analysis | |
CN103208185A (en) | Method and system for nighttime vehicle detection on basis of vehicle light identification | |
Li et al. | Robust people counting in video surveillance: Dataset and system | |
CN111640136B (en) | Depth target tracking method in complex environment | |
Zhang et al. | Vehicle recognition algorithm based on Haar-like features and improved Adaboost classifier | |
CN104615986A (en) | Method for utilizing multiple detectors to conduct pedestrian detection on video images of scene change | |
CN110163109B (en) | Lane line marking method and device | |
CN103679205A (en) | Preceding car detection method based on shadow hypothesis and layered HOG (histogram of oriented gradient) symmetric characteristic verification | |
Chao et al. | Multi-lane detection based on deep convolutional neural network | |
Naufal et al. | Preprocessed mask RCNN for parking space detection in smart parking systems | |
CN104112138A (en) | Object color classification method and device | |
CN111127520A (en) | Vehicle tracking method and system based on video analysis | |
CN115424217A (en) | AI vision-based intelligent vehicle identification method and device and electronic equipment | |
CN110909656A (en) | Pedestrian detection method and system with integration of radar and camera | |
Zhang et al. | Vehicle detection in UAV aerial images based on improved YOLOv3 | |
Zhang et al. | A front vehicle detection algorithm for intelligent vehicle based on improved gabor filter and SVM |
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 |