CN102646199B - Motorcycle type identifying method in complex scene - Google Patents
Motorcycle type identifying method in complex scene Download PDFInfo
- Publication number
- CN102646199B CN102646199B CN 201210049730 CN201210049730A CN102646199B CN 102646199 B CN102646199 B CN 102646199B CN 201210049730 CN201210049730 CN 201210049730 CN 201210049730 A CN201210049730 A CN 201210049730A CN 102646199 B CN102646199 B CN 102646199B
- Authority
- CN
- China
- Prior art keywords
- parts
- vehicle
- video image
- score
- search tree
- 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.)
- Expired - Fee Related
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a motorcycle type identifying method in a complex scene, which comprises the following steps of: initializing a component dictionary of a video image, learning a parameter of each component in the component dictionary, calculating an optimal composition structure according to the learned parameters of the components and an XOR searching tree, training and integrating vehicle templates by adopting the optimal composition structure, and detecting and identifying a motorcycle type in the video image by using the vehicle templates. According to the motorcycle type identifying method, the optimal composition structure of the vehicle templates is learned by adopting a dynamic planning algorithm, the XOR searching tree and a large quantity of actual samples, thus the efficiency of training the templates is increased, better discrimination is achieved, and actual application is facilitated. According to the motorcycle type identifying method, by combining a Latent SVM (Support Vector Machine) algorithm and a robust HOG (Histograms of Oriented Gradients) characteristic, the motorcycle type in the complex scene can be processed, and instantaneity and generality are ensured.
Description
Technical field
The present invention relates to image model identification, intelligent video monitoring and intelligent transportation field, be specifically related to the model recognizing method in a kind of complex scene.
Background technology
Vehicle identification based on video image refers to automatically identify dissimilar car from image and video, as minibus, car, truck, motor bus etc., it is the gordian technique in intelligent transportation system, no matter in the intelligent traffic monitoring field, or, in the full automatic charging field in highway and parking lot, it has extremely important application.
Vehicle identification based on video image generally is divided into three parts: 1, vehicle image cuts apart; 2, feature extraction; 3, the identification of vehicle and classification.In document, relevant model recognizing method mainly comprises at present: (a) model recognizing method based on prototype and (b) model recognizing method based on classification.
For the method based on prototype, often need the template database of model standard, then the vehicle image after over-segmentation, feature extraction and the template in database are mated.It generally can be divided into: the i) coupling based on vehicle edge; Ii) coupling based on vehicle ' s contour; Iii) coupling based on vehicle geometric parameter (as height, width, length and length breadth ratio etc.).These class methods are simple, intuitive the most, but its shortcoming is also quite obvious: one, and from real image, edge, profile or other geometric parameters of accurate extraction vehicle are more difficult; Its two, this method often requires video camera must be arranged on fixing position and, to its demarcation, has limited its application scenario; They are three years old, this method generally can only be separated size, length breadth ratio differs apparent in view vehicle, as large car and compact car, and differ unconspicuous car (for example be all lorry and the passenger vehicle of large car, or be all car and the just very difficult differentiation of jeep of compact car) for size, length breadth ratio; Its four, the inadequate robust of this method, be easy to be subject to the impact of picture noise, weather condition.
Method for based on classification, often need at first vehicle to be extracted to various features, then the sorter reasonable in design vehicle of classifying.The performance of these class methods often depends on the selection of feature and the design of sorter.It generally can be divided into: the i) identification of the vehicle based on neural network; Ii) identification of the vehicle based on the Gabor wave filter; Iii) identification of the vehicle based on support vector machine (SVM).Wherein, i) using the parameter of the 3 d structure model of vehicle as feature, then utilize neural network to be classified to the type of vehicle, ii) extracted the Gabor feature of vehicle, then utilize the method for template matches to realize vehicle classification, iii) extract some feature (as absolute altitude, width and the length of vehicle, SIFT feature etc.) of vehicle, then utilized support vector machine (SVM) to vehicle classification.Although these class methods, than the method based on prototype, have stronger robustness, they also exist common shortcoming: one, and these class methods still depend on the quality that image is cut apart very much, often can only process the simple situation of background; Its two, the feature that this class methods are selected or robust not, they are three years old, the model that these class methods adopt is all fairly simple, the coarse information that can only mean target, generally also can only be divided into vehicle large, medium and small three types, and can not carry out further exhaustive division; Its four, these class methods still highly depend on the placement location of video camera.
Recently, objective classification method based on parts has become a kind of trend, especially the partial model based on deformation (Deformable part template) that Felzenswalb proposes has been obtained great success and (has been seen " Object Detection with Discriminatively Trained Part Based Models ", IEEE Transactions on Pattern Analysis and Machine Intelligence, 32 (9): 1627-1645,2010).Two-layer Star Model of Latent-SVM Algorithm for Training for the method, this models coupling the geometry site between whole object and target component, have following advantage than recognition methods before: (1) has been used the more HOG feature of robust, make model have more identification, effectively overcome the sensitivity of existing method to complicated applications background and noise; (2) adopt the partial model based on deformation, allow parts to change on certain direction, position and yardstick, detailed information that therefore can target acquisition.Traditional recognition methods during vehicle, is often made car as a whole identification in identification, and this model is not only according to car load, and the parts of also waiting for bus according to wheel, vehicle window are identified car, have increased the reliability of identification; (3) this method does not need in advance target to be cut apart, and has therefore avoided classic method due to inaccurate the brought difficulty of Target Segmentation.Yet, because the method use heuristic carries out initialization to parts, the initialized location that therefore can not always find for parts, this tends to cause the inefficacy of model; Secondly, this heuristic also highly depends on number and the shape of the parts of artificial setting.In practice, components number and the shape of target are not often fixed, and it depends on the distance of video camera visual angle, distance objective and the difference between the target classification, and like this, for the vehicle of every type, one group of appropriate parts is selected in very difficult artificially.
Summary of the invention
The object of the present invention is to provide the model recognizing method in a kind of complex scene, it can locate and identify the type of vehicle efficiently, and greatly improves the speed of vehicle identification.
The present invention is achieved by the following technical solutions:
Model recognizing method under a kind of complex scene comprises the following steps:
(1) the parts dictionary of initialization video image comprises following sub-step:
(1-1) determine length breadth ratio and the area of detection window in video image according to positive and negative size in video image;
(1-2) according to the shape of parts in the length breadth ratio of detection window and area definition parts dictionary, area and point of fixity;
(1-3) according to the shape of parts, area and point of fixity build with or search tree;
(2) parameter of each parts in study parts dictionary;
(3) according to the parameter of each parts of study and with or search tree calculate the optimum structure that forms, comprise following sub-step:
(3-1) score at positive negative sample according to each parts of calculation of parameter of each parts of study,
And initialization and or the leaf node of search tree;
(3-2) according to the score of positive negative sample calculate with or the top score of search tree;
(3-3) determine selected node according to top score on and/or tree, to obtain the optimum structure that forms;
(4) adopt optimum structured training and the integration car modal of forming;
(5) use the vehicle in car modal detection and Identification video image.
In step (2), it is the parameter of using each parts in Latent-SVM Algorithm Learning parts dictionary.
In step (3-2), be by dynamic programming algorithm calculate from bottom to top with or the top score of search tree.
In step (3-3), be by the optimum structure that forms of retrogressive method calculating from bottom to top.
Step (4) specifically comprises: different angles, dissimilar vehicle are carried out training and the integration of template, and the threshold value of different templates is unitized.
Step (5) specifically comprises: adopt the method for moving window when the detection and Identification vehicle, and video image is extracted to HOG feature pyramid.
With respect to prior art, the present invention has following advantage and beneficial effect:
(1) car modal in the present invention utilized from a large amount of training sample learnings to the optimum structure that forms of parts, effectively improved the identification of template and the accuracy rate of identification;
(2) vehicle detection in the present invention has adopted the moving window method, has effectively overcome existing background subtraction method, frame-to-frame differences method and the optical flow method sensitivity to noise, has effectively overcome the impact of picture noise, has expanded widely the range of application of the method;
(3) the present invention combines the HOG feature of Latent SVM algorithm and robust, and the component model that trains corresponding composition structure according to type and the visual angle of car, do not need video camera to fix, can process the vehicle identification under complex scene, guaranteed real-time and versatility;
(4) method of the present invention is not limited to rough vehicle detection and classification, and as compact car and large car, it can carry out more careful classification, such as car and jeep, taxi and minibus etc.
The accompanying drawing explanation
The process flow diagram that Fig. 1 is the model recognizing method in complex scene of the present invention.
Fig. 2 is training sample and corresponding template thereof.
Fig. 3 illustrate with or search tree.
Fig. 4 (a) illustrates the vehicle recognition result of the inventive method for car.
Fig. 4 (b) illustrates the vehicle recognition result of the inventive method for truck.
Embodiment
Below at first technical term of the present invention is explained and illustrated.
Parts: corresponding to the part of vehicle, may be the zone etc. together on wheel, vehicle window, car door or vehicle body;
Component dictionary: the set formed by all parts;
Positive sample: formed by two parts: the image that comprises vehicle and the vehicle position (with upper left corner coordinate and the lower right corner Labeling Coordinate of rectangle frame) in image.
Negative sample: the image that does not comprise vehicle.
Learning sample: i.e. positive negative sample.
With or search tree: a concept in artificial intelligence and computer vision, be by with or figure promote, with or figure be a kind of by PROBLEM DECOMPOSITION, be systematically minor issue independently mutually, then divide and the method that solves.With or figure in two kinds of representational nodes are arranged: " with node " and " or node "." with node " refers to all subsequent node there is while solution, and it just has solution; " or node " refers to that each subsequent node is all fully independent, as long as wherein there is one to have and separate it solution is just arranged.For and/or tree, except start node, all the other each nodes only have a father node.
Recall: mainly refer to after having tried to achieve optimum solution, start to the leaf node search from the root node of and/or tree, with determine in asking the optimum solution process node or the state of process.
Detection window: refer to a rectangle frame on image, when detecting target, with rectangle frame scan image on a plurality of yardsticks, each step of scanning only is concerned about the image information in rectangle frame, see in this rectangle frame and whether comprise target, this rectangle frame is exactly figuratively a window.
As shown in Figure 1, the model recognizing method in complex scene of the present invention is as follows:
1, the parts dictionary of initialization video image, specifically comprise following sub-step
(1-1) determine length breadth ratio and the area of detection window in video image according to positive and negative size in video image; To one group of training sample D={x
1, x
2..., x
n... }, use the length breadth ratio of the peak value of sample length breadth ratio and Gaussian function convolution as detection block, use sample area 20 percent to divide for the area of a little making the position detection block;
(1-2) according to the shape of parts in the length breadth ratio of detection window and area definition parts dictionary, area and point of fixity;
Particularly, according to the size of detection window, enumerate area, length breadth ratio and the point of fixity of all candidates, wherein, the area of parts can not be greater than half of detection block area.The edge of parts can not surpass the edge of detection block.
(1-3) according to the shape of parts, area and point of fixity build with or search tree;
As shown in Figure 2, wherein with the node representative, the parts in father node are split as to two subassemblies, or node represents different fractionation modes.With or search tree enumerate the component relationship of all parts, a kind of composition structure of each stalk tree corresponding component.In addition, the candidate in dictionary meets the size that size is no less than 3 * 3 HOG piece and is not more than thick yardstick template.The template of training sample and correspondence thereof as shown in Figure 1.
2, use the parameter of each parts in Latent-SVM Algorithm Learning parts dictionary;
Choose candidates all in dictionary, its parameter can be obtained by the Latent-SVM Algorithm Learning:
Here, w is the long vector that the parameter of all M parts forms, w
jit is the parameter of j parts.σ
j(x
i, h
j) be the HOG feature that j parts extract.H
jbe hidden variable, specifically represent the position of the feature that each parts extracts, and the anglec of rotation.
3, according to the parameter of each parts of study and with or search tree calculate the optimum structure that forms;
Particularly, from dictionary, select one group not overlapping and cover the parts of detection window fully.Whether each parts is selected is must assign to determine on all positive negative samples according to it, and this score is calculated by following formula:
According to the score of each candidate, by dynamic programming algorithm with or search tree on calculate optimum composition structure, as shown in Figure 3, specifically comprise following sub-step:
(3-1) score at positive negative sample according to each parts of calculation of parameter of each parts of study, and initialization and or the leaf node of search tree.The score of each candidate calculated according to formula (3), and be assigned to or search tree in corresponding each leaf node.Other leaf node is composed and is divided into 0;
(3-2) according to the score of positive negative sample, use dynamic programming algorithm is from bottom to top calculated top score.According to the score of each leaf node, can calculate the top score of each node.The score of each and node be two leaf nodes branch and, each or node score are the maximal values of all child node scores;
(3-3) recall whole and/or tree from root node to leaf node according to top score, determine selected node, thereby obtain the optimum structure that forms.According to the method for recalling, can obtain optimal path, the parts that comprise in this optimal path are the optimum parts that form in structure.Selected parts have formed optimum composition structure;
4, adopt optimum structured training and the integration car modal of forming;
Particularly, the optimum structure that forms of the parts that utilize step (3) learning to arrive, we carry out the training of template to different angles, dissimilar vehicle, for example, in order to identify these two kinds of vehicles of car and truck, we may need car and three kinds of visual angles of truck Further Division: headstock, the tailstock and car are leaned to one side.We just need 6 vehicle templates of training like this, and final car modal has just comprised this 6 templates.For the detection threshold of unitized each template, we also need to adjust the bias term between each template in addition, and the threshold value of final template and the bias term of each template all arrive at the training sample learning by the Latent-SVM algorithm.
On training sample, the position of each parts and size are not in advance demarcated, and belong to hidden variable, so the training need of template adopts the coordinate descent algorithm, and the coordinate descent algorithm is divided into two steps: the 1) parameter of fixed form, locate the position of each parts; 2) position of fixed part, the parameter of learning template.Algorithm is iteration between these two processes always, until meet end condition.Simultaneously, for the convergence of accelerating algorithm, we have adopted the technology of data mining difficulty negative sample, when iteration training each time, dynamically add the difficult negative sample that classification makes mistakes, and dynamically remove the simple negative sample away from classifying face.
5, use the vehicle in car modal detection and Identification video image.
Detection and Identification to vehicle adopt the moving window method, idiographic flow is as shown in Fig. 4 (a) and Fig. 4 (b), for the two field picture in video flowing, at first we extract HOG feature pyramid on a plurality of yardsticks, then the template of utilizing the 4th step training to obtain, on the feature pyramid, detect successively and recognition image in the vehicle that comprises, this process is exactly the response of calculating car modal and HOG proper vector, if response is higher than the threshold value detected, algorithm is just predicted and a car detected here so.Wherein, for each candidate's vehicle, its corresponding vehicle is exactly the corresponding vehicle of template with peak response.For example, we are by a car be comprised of 6 vehicle templates (Car) and truck (Truck) integrated template.For the car detected in image, algorithm contrasts the response of various types of car modals, if the response maximum of the tailstock template of truck, this car detected so is exactly truck.For each car in image, algorithm is exported position and the corresponding vehicle classification at its place, as shown in Fig. 5 a and Fig. 5 b.
In addition, when the detection and Identification vehicle, we adopt Cascade beta pruning algorithm, go out a series of parts pruning threshold at the training sample learning, so original testing process has been divided into to a plurality of stages, can carry out parallel detection and identification to the car modal of a plurality of vehicles and angle, greatly improve the travelling speed of algorithm.
Claims (4)
1. the model recognizing method under a complex scene, is characterized in that, comprises the following steps:
(1) the parts dictionary of initialization video image comprises following sub-step:
(1-1) determine length breadth ratio and the area of detection window in described video image according to positive and negative size in described video image;
(1-2) according to the shape of parts in the length breadth ratio of described detection window and the described parts dictionary of area definition, area and point of fixity;
(1-3) according to the shape of described parts, area and point of fixity build with or search tree;
(2) use the parameter of each parts in the described parts dictionary of Latent-SVM Algorithm Learning;
(3) according to the parameter of each parts of described study and described and or search tree calculate the optimum structure that forms, comprise following sub-step:
(3-1) score at described positive negative sample according to each parts of calculation of parameter of each parts of described study, and initialization described with or the leaf node of search tree;
(3-2) according to the score of described positive negative sample calculate described with or the top score of search tree;
(3-3) determine selected node according to described top score on described and/or tree, to obtain the described optimum structure that forms;
(4) adopt described optimum structured training and the integration car modal of forming;
(5) use the vehicle in the described video image of described car modal detection and Identification, specifically comprise: adopt the method for moving window when the detection and Identification vehicle, and described video image is extracted to HOG feature pyramid.
2. the model recognizing method under a kind of complex scene according to claim 1, is characterized in that, in described step (3-2), be by dynamic programming algorithm calculate from bottom to top described with or the top score of search tree.
3. model recognizing method according to claim 1, is characterized in that, in described step (3-3), is by the described optimum structure that forms of retrogressive method calculating from bottom to top.
4. model recognizing method according to claim 1, is characterized in that, described step (4) specifically comprises: different angles, dissimilar vehicle are carried out training and the integration of template, and the threshold value of different templates is unitized.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 201210049730 CN102646199B (en) | 2012-02-29 | 2012-02-29 | Motorcycle type identifying method in complex scene |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 201210049730 CN102646199B (en) | 2012-02-29 | 2012-02-29 | Motorcycle type identifying method in complex scene |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102646199A CN102646199A (en) | 2012-08-22 |
| CN102646199B true CN102646199B (en) | 2013-12-18 |
Family
ID=46659015
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 201210049730 Expired - Fee Related CN102646199B (en) | 2012-02-29 | 2012-02-29 | Motorcycle type identifying method in complex scene |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102646199B (en) |
Families Citing this family (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102930271A (en) * | 2012-09-21 | 2013-02-13 | 博康智能网络科技股份有限公司 | Method for identifying taxicabs in real time by utilizing video images |
| CN102902983B (en) * | 2012-09-21 | 2016-01-20 | 博康智能网络科技股份有限公司 | A kind of taxi identification method based on support vector machine |
| CN102880881A (en) * | 2012-09-25 | 2013-01-16 | 常州大学 | Method for identifying car type on basis of binary support vector machines and genetic algorithm |
| CN102902960B (en) * | 2012-09-25 | 2015-04-22 | 武汉烽火众智数字技术有限责任公司 | Leave-behind object detection method based on Gaussian modelling and target contour |
| CN103942558A (en) * | 2013-01-22 | 2014-07-23 | 日电（中国）有限公司 | Method and apparatus for obtaining object detectors |
| CN103093249B (en) * | 2013-01-28 | 2016-03-02 | 中国科学院自动化研究所 | A kind of taxi identification method based on HD video and system |
| CN103258213B (en) * | 2013-04-22 | 2016-04-27 | 中国石油大学(华东) | A kind of for the dynamic vehicle model recognizing method in intelligent transportation system |
| CN103473566B (en) * | 2013-08-27 | 2016-09-14 | 东莞中国科学院云计算产业技术创新与育成中心 | A kind of vehicle checking method based on multiple dimensioned model |
| CN103886583A (en) * | 2014-02-14 | 2014-06-25 | 杭州电子科技大学 | Target detection sliding window scanning method based on scene geometric constraint |
| CN104091151A (en) * | 2014-06-30 | 2014-10-08 | 南京信息工程大学 | Vehicle identification method based on Gabor feature extraction and sparse representation |
| CN104751190B (en) * | 2015-04-23 | 2017-12-29 | 武汉大学 | A kind of vehicle part localization method finely identified towards vehicle |
| CN108475471B (en) * | 2016-01-26 | 2020-12-18 | 三菱电机株式会社 | Vehicle determination device, vehicle determination method, and computer-readable recording medium |
| CN105975949A (en) * | 2016-05-26 | 2016-09-28 | 大连理工大学 | Visual-information-based automobile identification method |
| CN106056102A (en) * | 2016-07-01 | 2016-10-26 | 哈尔滨工业大学 | Video-image-analysis-based road vehicle type classification method |
| CN107122706A (en) * | 2017-03-17 | 2017-09-01 | 华南理工大学 | A kind of vehicle checking method based on growth form extreme learning machine |
| CN107194338A (en) * | 2017-05-14 | 2017-09-22 | 北京工业大学 | Traffic environment pedestrian detection method based on human body tree graph model |
| CN108154127A (en) * | 2017-12-27 | 2018-06-12 | 天津智芯视界科技有限公司 | A kind of vehicle identification method based on video and radar |
| CN109902591B (en) * | 2018-03-13 | 2023-10-27 | 北京影谱科技股份有限公司 | Automobile searching system |
| CN108647624A (en) * | 2018-05-04 | 2018-10-12 | 广东工业大学 | A kind of road vehicle detection method, device, equipment and storage medium |
| CN109544907A (en) * | 2018-10-23 | 2019-03-29 | 大唐软件技术股份有限公司 | A kind of vehicle count method, device |
| CN110245670B (en) * | 2019-06-11 | 2021-08-20 | 观博云标(北京)文化科技有限公司 | Image pyramid gradient histogram feature-based skyhead identity identification method and device |
| CN112255973A (en) * | 2019-07-02 | 2021-01-22 | 库卡机器人(广东)有限公司 | Target detection method, detection terminal and storage medium in industrial production system |
| CN110765929A (en) * | 2019-10-21 | 2020-02-07 | 东软睿驰汽车技术（沈阳）有限公司 | Vehicle obstacle detection method and device |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN100470581C (en) * | 2005-12-29 | 2009-03-18 | 电子科技大学 | Vehicle recognition algorithm based on contour |
| CN101872477B (en) * | 2009-04-24 | 2014-07-16 | 索尼株式会社 | Method and device for detecting object in image and system containing device |
| CN101661559B (en) * | 2009-09-16 | 2013-03-06 | 中国科学院计算技术研究所 | Digital image training and detecting methods |
-
2012
- 2012-02-29 CN CN 201210049730 patent/CN102646199B/en not_active Expired - Fee Related
Also Published As
| Publication number | Publication date |
|---|---|
| CN102646199A (en) | 2012-08-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102646199B (en) | Motorcycle type identifying method in complex scene | |
| CN110942000B (en) | Unmanned vehicle target detection method based on deep learning | |
| CN105160309B (en) | Three lanes detection method based on morphological image segmentation and region growing | |
| CN102663356B (en) | Method for extraction and deviation warning of lane line | |
| Weber et al. | DeepTLR: A single deep convolutional network for detection and classification of traffic lights | |
| CN111563412B (en) | Rapid lane line detection method based on parameter space voting and Bessel fitting | |
| CN101859382B (en) | License plate detection and identification method based on maximum stable extremal region | |
| Behley et al. | Laser-based segment classification using a mixture of bag-of-words | |
| CN104392212B (en) | The road information detection and front vehicles recognition methods of a kind of view-based access control model | |
| Sivaraman et al. | Vehicle detection by independent parts for urban driver assistance | |
| CN103034836B (en) | Road sign detection method and road sign checkout equipment | |
| CN105892471A (en) | Automatic automobile driving method and device | |
| CN101980245B (en) | Adaptive template matching-based passenger flow statistical method | |
| CN101996401A (en) | Target analysis method and device based on intensity image and range image | |
| CN114359876B (en) | Vehicle target identification method and storage medium | |
| CN104615986A (en) | Method for utilizing multiple detectors to conduct pedestrian detection on video images of scene change | |
| Zhang et al. | A framework for turning behavior classification at intersections using 3D LIDAR | |
| CN107274668A (en) | A kind of congestion in road modeling method based on vehicle detection | |
| Kiy | A new method of global image analysis and its application in understanding road scenes | |
| CN109543498B (en) | Lane line detection method based on multitask network | |
| Kim et al. | Semantic point cloud-based adaptive multiple object detection and tracking for autonomous vehicles | |
| CN103295003A (en) | Vehicle detection method based on multi-feature fusion | |
| CN102945602A (en) | Vehicle trajectory classifying method for detecting traffic incidents | |
| CN115685237A (en) | Multi-mode three-dimensional target detection method and system combining viewing cones and geometric constraints | |
| Chen et al. | Research on vehicle detection and tracking algorithm for intelligent driving |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20131218 Termination date: 20160229 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |