CN107704833A - A kind of front vehicles detection and tracking based on machine learning - Google Patents
A kind of front vehicles detection and tracking based on machine learning Download PDFInfo
- Publication number
- CN107704833A CN107704833A CN201710952378.5A CN201710952378A CN107704833A CN 107704833 A CN107704833 A CN 107704833A CN 201710952378 A CN201710952378 A CN 201710952378A CN 107704833 A CN107704833 A CN 107704833A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- feature
- tracking
- detection
- lbp
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/285—Selection of pattern recognition techniques, e.g. of classifiers in a multi-classifier system
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
Abstract
The present invention relates to a kind of front vehicles detection based on machine learning and tracking.The present invention includes two processes of vehicle detection and vehicle tracking.In the vehicle detection stage, the vehicle in front of road surface is substantially detected using LBP features and Adaboost classifier trainings result；These positions are verified to eliminate non-vehicle target using the training result of HAAR features and HOG features again.The vehicle tracking stage, using improved core correlation filtering.The present invention has the characteristics of high verification and measurement ratio, tracking efficiency high.
Description
Technical field
The invention belongs to image processing and pattern recognition field, particularly a kind of front vehicles based on machine learning
Detection and tracking, for the front vehicles target in Vehicular video, detect and track its movement locus.
Background technology
Exploitation vehicle carried driving accessory system (driver assistance system) is it should be understood that to be related to vehicle attached
Shortcut road vehicles moving situation, it is DAS key message to determine position of other vehicles on road.Therefore, strong robustness
Vehicle detection and vehicle tracking research are crucial.
At present, researcher is based on radar or laser radar by using such as having chance with sensor to solve this problem,
However, the research of vehicle detection and tracking based on monocular vision has in terms of cost and maintenance cost and information fidelity is reduced
There is preferable advantage.But in highway and urban transportation, only carrying out detect and track front vehicles using only monocular camera is still
One challenging task.Major Difficulties have the following aspects：First, with the installation of vehicle-mounted camera, vehicle inspection
Method of determining and calculating is faced with the background of change and the great variety of lighting environment.Secondly, various vehicles are handed in highway and city
There are different shape and color in logical, the big modeling caused to vehicle of quantity is very difficult.3rd, itself car on road
And front vehicles be in motion state, and in this way, the size and location of vehicle in image space is a variety of
Various.
The method of many vehicle detections is widely used：(1) method based on known knowledge；(2) feature matching method；
(3) sensor fusion method.Method based on known knowledge is come identification object using vehicle feature in itself, and is assumed
Potential vehicle location, such as cable architecture, vehicle shadow, image symmetrical, etc..Research focuses principally on the single spy of vehicle above
The detection of sign, also both it is possible to that there are these features in image background, a large amount of flase drops can be caused.
The content of the invention
In view of the shortcomings of the prior art, the present invention proposes a kind of front vehicles detection based on machine learning and track side
Method.
The technical solution adopted for solving the technical problem of the present invention is：
Step 1, the video information for obtaining real-time vehicle.
Step 2, the frame of video to acquisition carry out image preprocessing, image gray processing, extract the region of vehicle target.
Step 3, using priori, according to general scene and the visual field of camera, intercept area-of-interest, remove day
The interference of empty, peripheral part building and billboard mark.
Step 4, using the grader of LBP features trainings target detection is carried out to area-of-interest, obtained comprising vehicle
Candidate region.
4.1st, vehicle LBP feature extractions
In 3*3 window, using center single pixel as threshold value, compare the size of surrounding pixel and its gray value, if than
It is big, then pixel of changing the time is flagged as 1, is otherwise 0；8 binary systems can be drawn, obtain the window center pixel
LBP values, it is LBP values to be converted into the decimal system.
4.2nd, sorter model is trained
First, using the picture of only vehicle as positive sample, the picture including vehicle is not as negative sample, and by it all
Described with LBP operators；Then trained to obtain different Weak Classifiers with GentleAdaboost sorter model, and will be by one
Fixed weighted superposition gets up, the strong classifier that composition is finally wanted；Finally, using sequence of video images as input, categorized device
Detection obtains candidate's vehicle.
Step 5, using the grader of HAAR features trainings and the grader of HOG features trainings to the above-mentioned time for including vehicle
Favored area carries out target detection, rejects non-vehicle region.
5.1st, vehicle HAAR feature extractions
Feature templates are formed using edge feature, linear character, central feature or diagonal feature, and use it to calculate sample
This HAAR characteristic-integration figures.
5.2nd, vehicle HOG feature extractions
Sample image is divided into small connected region, then gathers the gradient of wherein each pixel or the direction Histogram at edge
Figure, retraining obtain Adaboost feature classifiers.
Step 6, after obtaining vehicle detection target, using improved core correlation filtering, vehicle target is tracked, and often
Compare testing result every 10 frames and whether tracking result is correct, more than threshold range, then tracking result is carried out with testing result
Amendment.
The beneficial effects of the invention are as follows：The present invention has and is influenceed that smaller, cost is relatively low and flase drop missing inspection by illumination variation
The low advantage of rate.
Brief description of the drawings
Fig. 1 is cascade classifier training and detection process based on LBP feature learnings；
Fig. 2 is cascade classifier training and detection process based on HAAR feature learnings；
Fig. 3 is cascade classifier training and detection process based on HOG feature learnings；
Fig. 4 is tracking and testing process；
Fig. 5 is car test survey of the present invention and trace flow figure.
Embodiment
Below in conjunction with accompanying drawing 5, the invention will be further described.
The present invention comprises the following steps：
Step 1, the video information for obtaining real-time vehicle；
Step 2, the frame of video to acquisition carry out image preprocessing, image gray processing, extract the region of vehicle target；
Step 3, using priori, according to general scene and the visual field of camera, intercept area-of-interest, remove day
Empty, peripheral part building and billboard mark etc. interference；
Step 4, using the grader of LBP features trainings target detection is carried out to area-of-interest, obtained comprising vehicle
Candidate region；
4.1st, vehicle LBP feature extractions
LBP is a kind of operator for being used for describing image local textural characteristics, and it has rotational invariance and gray scale consistency
Etc. significant feature.LBP operator definitions are in 3*3 window, using center single pixel as threshold value, compare surrounding pixel and its
The size of gray value, if bigger than its, pixel of changing the time is flagged as 1, is otherwise 0.Therefore 8 binary systems can be drawn, obtained
To the LBP values of the window center pixel, it is LBP values to be converted into the decimal system.
4.2nd, sorter model is trained
First, using the picture of only vehicle as positive sample, the picture including vehicle is not as negative sample, and by it all
Described with LBP operators.Then trained to obtain different Weak Classifiers with GentleAdaboost sorter model, and will be by one
Fixed weighted superposition gets up, the strong classifier that composition is finally wanted.Finally, using sequence of video images as input, categorized device
Detection obtains candidate's vehicle.As shown in Figure 1.
Step 5, using the grader of HAAR features trainings and the grader of HOG features trainings to the above-mentioned time for including vehicle
Favored area carries out target detection, rejects non-vehicle region, improves accuracy rate；
5.1st, vehicle HAAR feature extractions
HAAR operators are to form spy with specific feature, such as edge feature, linear character, central feature and diagonal feature
Template is levied, and uses it to calculate the HAAR characteristic-integration figures of sample.Detailed process is as shown in Figure 2.
5.2nd, vehicle HOG feature extractions
HOG operators are that sample image is divided into small connected region, then gather the gradient or edge of wherein each pixel
Direction histogram, retraining obtains Adaboost feature classifiers.Detailed process is as shown in Figure 3.
Step 6, after obtaining vehicle detection target, using improved core correlation filtering (KCF) algorithm, vehicle target is tracked,
And compare testing result every 10 frames and whether tracking result is correct, more than threshold range, then tracking is tied with testing result
Fruit is modified.
Once it is that testing result obtains some vehicle that the vehicle tracking algorithm in the present invention, which is briefly described, the algorithm can be
Around it respectively upwards, move down different pixels and obtain new sample image, directly increase the quantity of sample, Ran Houxun
Get grader, you can detect vehicle.The correlation filter as two is devised in the present invention, for realizing mesh respectively
Mark tracking and change of scale.It is respectively defined as position filtering device (translation filter) and scaling filter (scale
filter).Position filtering device is used for positioning the target in present frame, and scaling filter is used for estimating the target scale in present frame
Size.Two wave filters are separate, so as to selecting different feature species and feature calculation mode training and
Detection.
First one group of gray level image f1, f2 ..., ft are extracted as training sample from tracking target, and by corresponding to it
Wave filter output is set to g1, g2 ..., gt, and optimal correlation filter needs to meet following condition[18]：
ε minimum value can be solved by following formula：
In general, gjCan be the output of arbitrary shape.In order to facilitate calculating, it is assumed herein that the g of outputjIt is Gaussian function
Number, its peak value are located at fjCenter.
G=h*f (3)
It can be obtained by Fourier transformation
It can obtain correlation filter H.
First in n-th frame, with the target detected in vehicle detection, extraction HOG features are trained as input.So
Assume that output is Gauss type function afterwards, then can is back-calculated to obtain correlation filter H.Then HOG features are extracted in the (n+1)th frame
As input, obtain exporting y accordingly by correlation filter.In general algorithm comparison is slow in calculating process, uses Cyclic Moment
Battle array and IFFT accelerate arithmetic speed.
Such as Fig. 4, part of detecting is to input Z using the image of next frame as test, and phase is can obtain by correlation filter H
The output y answered.IFFT is used during and, is converted to frequency domain, may be such that many is accelerated in calculating, this is also that of the invention one is big
Advantage.
Invention describes a kind of moving vehicles detection and tracking method based on machine learning, including vehicle detection and car
Tracking two processes.In the vehicle detection stage, outlet is substantially detected using LBP features and Adaboost classifier training results
Square vehicle in front；These positions are verified to eliminate non-vehicle target using the training result of HAAR features and HOG features again.
The vehicle tracking stage, using improved core correlation filtering (KCF) algorithm.Result of the test shows, in the case of different weather, is based on
The moving vehicles detection and tracking method of this algorithm has verification and measurement ratio height, tracking efficiency high, can reach the requirement detected in real time.Vehicle
Detect and track accuracy rate is more than 90%.Following table is the embodiment of the present invention and other method comparing result.
Varying environment ventrocephalad moving vehicles detection and tracking accuracy rate
Claims (1)
1. a kind of front vehicles detection and tracking based on machine learning, it is characterised in that this method comprises the following steps：
Step 1, the video information for obtaining real-time vehicle；
Step 2, the frame of video to acquisition carry out image preprocessing, image gray processing, extract the region of vehicle target；
Step 3, using priori, according to general scene and the visual field of camera, intercept area-of-interest, remove sky,
Peripheral part is built and the interference of billboard mark；
Step 4, using the grader of LBP features trainings target detection is carried out to area-of-interest, obtain including the candidate of vehicle
Region；
4.1st, vehicle LBP feature extractions
In 3*3 window, using center single pixel as threshold value, compare the size of surrounding pixel and its gray value, if bigger than its
, then pixel of changing the time is flagged as 1, is otherwise 0；8 binary systems can be drawn, obtain the LBP values of the window center pixel,
It is LBP values to be converted into the decimal system；
4.2nd, sorter model is trained
First, using the picture of only vehicle as positive sample, the picture including vehicle is not all used as negative sample, and by it
LBP operators describe；Then trained to obtain different Weak Classifiers with GentleAdaboost sorter model, and will be by certain
Weighted superposition get up, the final desired strong classifier of composition；Finally, using sequence of video images as input, categorized device inspection
Measure candidate's vehicle；
Step 5, using the grader of HAAR features trainings and the grader of HOG features trainings to the above-mentioned candidate regions for including vehicle
Domain carries out target detection, rejects non-vehicle region；
5.1st, vehicle HAAR feature extractions
Feature templates are formed using edge feature, linear character, central feature or diagonal feature, and use it to calculate sample
HAAR characteristic-integration figures；
5.2nd, vehicle HOG feature extractions
Sample image is divided into small connected region, then gathers the gradient of wherein each pixel or the direction histogram at edge,
Retraining obtains Adaboost feature classifiers；
Step 6, after obtaining vehicle detection target, using improved core correlation filtering, vehicle target is tracked, and every 10
Frame compares testing result and whether tracking result is correct, more than threshold range, then tracking result is modified with testing result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710952378.5A CN107704833A (en) | 2017-10-13 | 2017-10-13 | A kind of front vehicles detection and tracking based on machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710952378.5A CN107704833A (en) | 2017-10-13 | 2017-10-13 | A kind of front vehicles detection and tracking based on machine learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107704833A true CN107704833A (en) | 2018-02-16 |
Family
ID=61184221
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710952378.5A Pending CN107704833A (en) | 2017-10-13 | 2017-10-13 | A kind of front vehicles detection and tracking based on machine learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107704833A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109190523A (en) * | 2018-08-17 | 2019-01-11 | 武汉大学 | A kind of automobile detecting following method for early warning of view-based access control model |
| CN109492588A (en) * | 2018-11-12 | 2019-03-19 | 广西交通科学研究院有限公司 | A kind of rapid vehicle detection and classification method based on artificial intelligence |
| CN110517291A (en) * | 2019-08-27 | 2019-11-29 | 南京邮电大学 | A kind of road vehicle tracking based on multiple feature spaces fusion |
| CN110598758A (en) * | 2019-08-23 | 2019-12-20 | 伟龙金溢科技（深圳）有限公司 | Training modeling method, vehicle charging method, management system, and storage medium |
| CN110969065A (en) * | 2018-09-30 | 2020-04-07 | 北京四维图新科技股份有限公司 | Vehicle detection method and device, front vehicle anti-collision early warning equipment and storage medium |
| CN111008997A (en) * | 2019-12-18 | 2020-04-14 | 南京莱斯电子设备有限公司 | Vehicle detection and tracking integrated method |
-
2017
- 2017-10-13 CN CN201710952378.5A patent/CN107704833A/en active Pending
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109190523A (en) * | 2018-08-17 | 2019-01-11 | 武汉大学 | A kind of automobile detecting following method for early warning of view-based access control model |
| CN109190523B (en) * | 2018-08-17 | 2021-06-04 | 武汉大学 | Vehicle detection tracking early warning method based on vision |
| CN110969065A (en) * | 2018-09-30 | 2020-04-07 | 北京四维图新科技股份有限公司 | Vehicle detection method and device, front vehicle anti-collision early warning equipment and storage medium |
| CN110969065B (en) * | 2018-09-30 | 2023-11-28 | 北京四维图新科技股份有限公司 | Vehicle detection method and device, front vehicle anti-collision early warning device and storage medium |
| CN109492588A (en) * | 2018-11-12 | 2019-03-19 | 广西交通科学研究院有限公司 | A kind of rapid vehicle detection and classification method based on artificial intelligence |
| CN110598758A (en) * | 2019-08-23 | 2019-12-20 | 伟龙金溢科技（深圳）有限公司 | Training modeling method, vehicle charging method, management system, and storage medium |
| CN110517291A (en) * | 2019-08-27 | 2019-11-29 | 南京邮电大学 | A kind of road vehicle tracking based on multiple feature spaces fusion |
| CN111008997A (en) * | 2019-12-18 | 2020-04-14 | 南京莱斯电子设备有限公司 | Vehicle detection and tracking integrated method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107704833A (en) | A kind of front vehicles detection and tracking based on machine learning | |
| CN106683119B (en) | Moving vehicle detection method based on aerial video image | |
| CN104361343B (en) | Vehicle type recognition method and its device | |
| CN108154102B (en) | Road traffic sign identification method | |
| CN103927526B (en) | Vehicle detecting method based on Gauss difference multi-scale edge fusion | |
| Alefs et al. | Road sign detection from edge orientation histograms | |
| Kim et al. | Multi-task convolutional neural network system for license plate recognition | |
| Huang et al. | Vehicle detection and inter-vehicle distance estimation using single-lens video camera on urban/suburb roads | |
| CN109101924A (en) | A kind of pavement marking recognition methods based on machine learning | |
| CN108009518A (en) | A kind of stratification traffic mark recognition methods based on quick two points of convolutional neural networks | |
| CN105488454A (en) | Monocular vision based front vehicle detection and ranging method | |
| Zhang et al. | Study on traffic sign recognition by optimized Lenet-5 algorithm | |
| CN106815583B (en) | Method for positioning license plate of vehicle at night based on combination of MSER and SWT | |
| CN103903018A (en) | Method and system for positioning license plate in complex scene | |
| Zang et al. | Traffic sign detection based on cascaded convolutional neural networks | |
| CN107180230B (en) | Universal license plate recognition method | |
| Kim et al. | Effective traffic lights recognition method for real time driving assistance systemin the daytime | |
| CN104766344B (en) | Vehicle checking method based on movement edge extractor | |
| CN105893960A (en) | Road traffic sign detecting method based on phase symmetry | |
| Liu et al. | Multi-type road marking recognition using adaboost detection and extreme learning machine classification | |
| Umarov et al. | Traffic sign recognition method based on simplified Gabor wavelets and CNNs | |
| Chiang et al. | Road speed sign recognition using edge-voting principle and learning vector quantization network | |
| Phu et al. | Traffic sign recognition system using feature points | |
| CN108734170B (en) | License plate character segmentation method based on machine learning and template | |
| Das et al. | Automatic License Plate Recognition Technique using Convolutional Neural Network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20180216 |