CN104156697A - Vehicle type recognition method under night bayonet scene - Google Patents
Vehicle type recognition method under night bayonet scene Download PDFInfo
- Publication number
- CN104156697A CN104156697A CN201410355985.XA CN201410355985A CN104156697A CN 104156697 A CN104156697 A CN 104156697A CN 201410355985 A CN201410355985 A CN 201410355985A CN 104156697 A CN104156697 A CN 104156697A
- Authority
- CN
- China
- Prior art keywords
- car light
- car
- vehicle
- algorithm
- region
- 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
Landscapes
- Traffic Control Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a vehicle type recognition method under a night bayonet scene and application of the method in vehicle characteristic recognition. The method comprises the following steps: outputting license plate position and size based on license plate detection and recognition algorithm; extracting ROI of vehicle lamps according to the license plate position and size; extracting vehicle lamp foreground adopting self-adapting threshold value and area growing algorithm; creating multidimensional characteristic vector of the vehicle lamps and the area between the vehicle lamps utilizing vehicle lamp outline extracting algorithm and HOG characteristics; conducting vehicle type classification adopting SVM algorithm. The method effectively recognizes a vehicle type under a night bayonet scene, so as to enrich functions of the vehicle characteristic recognition system, and provide a more efficient vehicle management method for users of public security bureaus and the like.
Description
Technical field
The invention belongs to traffic computer vision field, the particularly model recognizing method under bayonet socket scene at a kind of night, and the application of the method in vehicle characteristics recognition system.
Technical background
Along with popularizing of high-definition video monitoring, public security becomes more and more convenient to the management of vehicle, what candid photograph was arrived crosses car data, except seeing clearly vehicle license, can also see vehicle producer mark, type of vehicle, body color, driver's facial characteristics even clearly, this solves a case and provides a great convenience public security officer.But car data is very huge owing to crossing, public security officer wants to search for target vehicle not a duck soup from mass data, need to expend a large amount of manpower and materials.Vehicle characteristics recognition technology based on computer vision technique can solve such problem very effectively, and vehicle characteristics, except the conventional number-plate number, car plate color, car plate producer mark, body color, also comprises type of vehicle.Comprehensive these conditions, can filter non-suspected vehicles quickly, improve the efficiency that public security is handled a case.Therefore, vehicle identification belongs to one of very important technology in vehicle characteristics recognition technology.
Because vehicle is all kinds of various, the standard that vehicle classification is ununified, according to the careful degree of classification, in general has three levels.Ground floor, can be divided into large, medium and small type car, the second layer, can be divided into compact car, minicar, compact vehicle, medium vehicle, senior vehicle, deluxe carmodel, three-box car type, MPV vehicle, SUV vehicle etc., the 3rd layer, be subdivided into the sub-brand name under certain producer's brand, as the Sagitar under popular brand, golf, Lang Yi, Magotan etc., the A4 of Audi, A6, the Q3 of Audi, Q7 etc. under Audi's brand.From the aspect of algorithm, ground floor classification is the easiest, and three-layer classification difficulty is quite large, but from user's angle, the quantity of information maximum that three-layer classification provides, the quantity of information minimum that ground floor provides, therefore they wish that algorithm can realize three-layer classification.
The invention provides model recognizing method under one bayonet socket scene at night, the method utilizes characteristics of image the binding pattern recognition methods in region between car light under scene at night and car light thereof to carry out vehicle classification, the method can be applicable to three-layer classification, has effectively solved a difficult problem for vehicle classification under scene at night.
Summary of the invention
The object of the invention is, in order to solve vehicle under bayonet socket scene at a night identification difficult problem, to become possibility thereby round-the-clock lower vehicle is identified.Summary of the invention is as follows:
(1), utilize car plate to detect and recognizer, output car plate position and size, and extract car light area-of-interest (ROI) according to car plate position and size, utilize the priori of car light and car plate position relationship, preliminary definite car light region, to improve car light extraction accuracy and efficiency.
(2), utilize adaptive threshold and algorithm of region growing to extract car light prospect, choose car light threshold value based on grey level histogram self-adaptation, extract car light center compared with bright area, then utilize the central point extracting as seed, carry out region growing, obtain complete car light region.
(3), utilize car light to occur in pairs, and priori based on car plate symmetry, carries out car light pairing, for the situation of only having single car light to occur, adopts and forces pairing strategy, matches by car light, has effectively reduced undetected side by side except certain flase drop.
(4), utilize algorithm of convex hull to calculate car light region salient point, and carry out matching, obtain accurate car light region, for feature extraction is below prepared.
(5), utilize car light profile extraction algorithm, extract profile length, area, Central Moment Feature, the Hu moment characteristics of car light, and based on the normalization of car plate size, as the feature of car light.
(6), utilize Hog feature, extract car light between region (this region often comprises abundant feature), and with car light Fusion Features, together as the feature of vehicle identification.
(7), utilize support vector machine (SVM) training classifier, because proper vector dimension is higher, and classification number is more, thus adopt linear SVM, to ensure that algorithm has good generalization.
Brief description of the drawings
Fig. 1 is basic flow sheet of the present invention;
Fig. 2 is car light foreground extraction algorithm flow chart of the present invention;
Fig. 3 is car light pairing algorithm flow chart of the present invention;
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is clearly and completely described, obviously, described embodiment is only the present invention's part embodiment, instead of whole embodiment.Based on the embodiment in the present invention, those of ordinary skill in the art, not making the every other embodiment obtaining under creative work prerequisite, belong to the scope of protection of the invention.
A model recognizing method under night bayonet socket scene, implementing procedure is as follows:
1, tentatively determine car light region.Utilize car plate to detect and recognizer, output car plate position and size, and extract car light area-of-interest (ROI) according to car plate position and size, utilize the priori of car light and car plate position relationship.In the embodiment of the present invention, getting left and right car light roi peak width is 2.5 times of car plate width, be highly 12.5 times of car plate height, the above region of car plate accounts for 2/3 of whole height, to reach in the situation of not missing car light, dwindle the effect of hunting zone as far as possible, improve car light extraction accuracy and efficiency.According to on-the-spot actual environment, can adjust above parameter, to reach best effect.
2, extract car light prospect.Utilize adaptive threshold and algorithm of region growing to extract car light prospect, calculate the grey level histogram in roi region, get pixel that 5% gray-scale value is the highest as threshold value, binaryzation obtains car light core pixel, with these pixel values as Seed Points, with eight neighborhood region growings, get complete car light region.
3, car light pairing.Based on three criterions: area is similar, on same horizontal line, based on car plate Central Symmetry, for three criterions, three fault-tolerant coefficients 0.5,0.5 and 0.2 is set respectively.For part, due to reflective too strong, whether the and bicycle lamp that falls that can not match, takes to force the strategy of pairing, based on car plate centralizing mapping, and according to dutycycle judgement, match correct.
4, car light reparation.The car light prospect extracting often exists some concave defects or fringe region exists a lot of burrs, noise etc., and the present invention utilizes algorithm of convex hull to calculate car light region salient point, and carries out matching, obtains accurate car light region.
5, extract car light feature.Utilize car light profile extraction algorithm, extract profile length, area, Central Moment Feature, the Hu moment characteristics of car light, and based on the normalization of car plate size.
6, extract car light between the feature of heat sink area.Utilize this feature of Hog feature extraction, and merge mutually with car light feature, as the feature of whole vehicle identification.
7, vehicle identification.Adopt SVM to carry out feature database training, because proper vector dimension is higher, and classification number is more, thus adopt linear SVM, to ensure that algorithm has good generalization.
Cannot be to a difficult problem of carrying out vehicle identification under bayonet socket scene night for classic method, the vehicle targets based on car light that the present invention proposes, has solved this problem effectively, and can be applicable in the vehicle identification of three-layer classification.
Claims (8)
1. the model recognizing method under a night bayonet socket scene, it is characterized in that utilizing car plate to detect and recognizer, output car plate position and size, and extract car light area-of-interest (ROI) according to car plate position and size, to improve car light extraction accuracy and efficiency; Utilize adaptive threshold and algorithm of region growing to extract car light prospect; Utilize the multidimensional characteristic vectors in region between car light profile extraction algorithm and HOG feature-modeling car light and car light; Utilize SVM algorithm to carry out vehicle classification.
2. method according to claim 1, it is characterized in that utilizing car plate to detect and recognizer, output car plate position and size, and extract car light area-of-interest (ROI) according to car plate position and size, utilize the priori of car light and car plate position relationship, preliminary definite car light region, to improve car light extraction accuracy and efficiency.
3. method according to claim 1, it is characterized in that utilizing adaptive threshold and algorithm of region growing to extract car light prospect, choose car light threshold value based on grey level histogram self-adaptation, extract car light center compared with bright area, then utilize the central point extracting as seed, carry out region growing, obtain complete car light region.
4. method according to claim 3, it is characterized in that utilizing car light to occur in pairs, and priori based on car plate symmetry, carry out car light pairing, for the situation of only having single car light to occur, adopt and force pairing strategy, match by car light, effectively reduced undetected side by side except certain flase drop.
5. method according to claim 3, is characterized in that utilizing algorithm of convex hull to calculate car light region salient point, and carries out matching, obtains accurate car light region, for feature extraction is below prepared.
6. method according to claim 1, is characterized in that utilizing car light profile extraction algorithm, extracts profile length, area, Central Moment Feature, the Hu moment characteristics of car light, and based on the normalization of car plate size, as the feature of car light.
7. method according to claim 1, is characterized in that utilizing Hog feature, extract car light between region (this region often comprises abundant feature), and the car light feature of comprehensive claim 6, together as the feature of vehicle identification.
8. method according to claim 1, it is characterized in that utilizing support vector machine (SVM) training classifier, because proper vector dimension is higher, and classification number is more, therefore employing linear SVM, to ensure that algorithm has good generalization.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410355985.XA CN104156697A (en) | 2014-07-24 | 2014-07-24 | Vehicle type recognition method under night bayonet scene |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410355985.XA CN104156697A (en) | 2014-07-24 | 2014-07-24 | Vehicle type recognition method under night bayonet scene |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN104156697A true CN104156697A (en) | 2014-11-19 |
Family
ID=51882194
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410355985.XA Pending CN104156697A (en) | 2014-07-24 | 2014-07-24 | Vehicle type recognition method under night bayonet scene |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104156697A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105095848A (en) * | 2014-12-01 | 2015-11-25 | 北京云识图信息技术有限公司 | Object identification method and system |
| CN106446929A (en) * | 2016-07-18 | 2017-02-22 | 浙江工商大学 | Vehicle type detection method based on edge gradient potential energy |
| CN106845493A (en) * | 2016-12-06 | 2017-06-13 | 西南交通大学 | The identification at railroad track close-range image rail edge and matching process |
| CN109359666A (en) * | 2018-09-07 | 2019-02-19 | 佳都新太科技股份有限公司 | A kind of model recognizing method and processing terminal based on multiple features fusion neural network |
| CN111243285A (en) * | 2020-01-07 | 2020-06-05 | 南京甄视智能科技有限公司 | Automatic vehicle fake plate detection method and system based on vehicle lamp image recognition in dark environment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1928892A (en) * | 2006-09-20 | 2007-03-14 | 王枚 | Method and device for license plate location recognition, vehicle-logo location recognition and vehicle type |
| CN103473534A (en) * | 2013-09-10 | 2013-12-25 | 西安翔迅科技有限责任公司 | Vehicle detecting method based on video |
| CN103942560A (en) * | 2014-01-24 | 2014-07-23 | 北京理工大学 | High-resolution video vehicle detection method in intelligent traffic monitoring system |
-
2014
- 2014-07-24 CN CN201410355985.XA patent/CN104156697A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1928892A (en) * | 2006-09-20 | 2007-03-14 | 王枚 | Method and device for license plate location recognition, vehicle-logo location recognition and vehicle type |
| CN103473534A (en) * | 2013-09-10 | 2013-12-25 | 西安翔迅科技有限责任公司 | Vehicle detecting method based on video |
| CN103942560A (en) * | 2014-01-24 | 2014-07-23 | 北京理工大学 | High-resolution video vehicle detection method in intelligent traffic monitoring system |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105095848A (en) * | 2014-12-01 | 2015-11-25 | 北京云识图信息技术有限公司 | Object identification method and system |
| CN106446929A (en) * | 2016-07-18 | 2017-02-22 | 浙江工商大学 | Vehicle type detection method based on edge gradient potential energy |
| CN106446929B (en) * | 2016-07-18 | 2019-02-22 | 浙江工商大学 | Type of vehicle detection method based on edge gradient potential energy |
| CN106845493A (en) * | 2016-12-06 | 2017-06-13 | 西南交通大学 | The identification at railroad track close-range image rail edge and matching process |
| CN109359666A (en) * | 2018-09-07 | 2019-02-19 | 佳都新太科技股份有限公司 | A kind of model recognizing method and processing terminal based on multiple features fusion neural network |
| CN111243285A (en) * | 2020-01-07 | 2020-06-05 | 南京甄视智能科技有限公司 | Automatic vehicle fake plate detection method and system based on vehicle lamp image recognition in dark environment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Braun et al. | Pose-rcnn: Joint object detection and pose estimation using 3d object proposals | |
| Wang et al. | A survey of vehicle re-identification based on deep learning | |
| Fritsch et al. | Monocular road terrain detection by combining visual and spatial information | |
| CN100485710C (en) | Method for recognizing vehicle type by digital picture processing technology | |
| US10025998B1 (en) | Object detection using candidate object alignment | |
| CN104850850B (en) | A kind of binocular stereo vision image characteristic extracting method of combination shape and color | |
| Mammeri et al. | Lane detection and tracking system based on the MSER algorithm, hough transform and kalman filter | |
| EP3343431A1 (en) | Method and system for vehicle localization from camera image | |
| WO2017041396A1 (en) | Driving lane data processing method, device, storage medium and apparatus | |
| CN104902261B (en) | Apparatus and method for the road surface identification in low definition video flowing | |
| CN104156697A (en) | Vehicle type recognition method under night bayonet scene | |
| Kim et al. | Robust detection of non-motorized road users using deep learning on optical and LIDAR data | |
| CN105740886B (en) | A kind of automobile logo identification method based on machine learning | |
| CN104299008A (en) | Vehicle type classification method based on multi-feature fusion | |
| CN108764096B (en) | Pedestrian re-identification system and method | |
| Shi et al. | A vision system for traffic sign detection and recognition | |
| Chang et al. | An efficient method for lane-mark extraction in complex conditions | |
| Vitor et al. | Comprehensive performance analysis of road detection algorithms using the common urban Kitti-road benchmark | |
| Mehtab et al. | 3D vehicle detection using cheap LiDAR and camera sensors | |
| Yang et al. | Vehicle detection from low quality aerial LIDAR data | |
| Lee et al. | independent object detection based on two-dimensional contours and three-dimensional sizes | |
| Schels et al. | Synthetically trained multi-view object class and viewpoint detection for advanced image retrieval | |
| Liu et al. | Obstacle recognition for ADAS using stereovision and snake models | |
| Schomerus et al. | Camera-based lane border detection in arbitrarily structured environments | |
| Boggavarapu et al. | A robust multi color lane marking detection approach for Indian scenario |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20141119 |
|
| WD01 | Invention patent application deemed withdrawn after publication |