CN105809167A - Method for parting vehicles sticking together in monitoring video - Google Patents
Method for parting vehicles sticking together in monitoring video Download PDFInfo
- Publication number
- CN105809167A CN105809167A CN201510109679.2A CN201510109679A CN105809167A CN 105809167 A CN105809167 A CN 105809167A CN 201510109679 A CN201510109679 A CN 201510109679A CN 105809167 A CN105809167 A CN 105809167A
- Authority
- CN
- China
- Prior art keywords
- adhesion
- vehicle
- point
- region
- vehicle 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.)
- Granted
Links
Abstract
The invention discloses a method for parting vehicles sticking together in a monitoring video. The method comprises the following steps: a step of vehicle area detection, a step of determining an area of the vehicles sticking together, a step of skeleton extraction at the area of the vehicles sticking together, a step of skeleton angle point detection and clustering, a step of determining parting lines, a step of parting the vehicles sticking together and the like. The method for parting the vehicles sticking together in the monitoring video helps solve a problem that vehicle tracking is affected because vehicles stick together in collected images in the field of vehicle tracking on highways.
Description
Technical field
The present invention relates to the technology of image processing field and computer realm, be a kind of for splitting the method that there is adhesion vehicle in monitor video concretely.
Background technology
In the moving vehicle based on video detects, monitoring camera robot the vehicle in the video image grasped has substantially three kinds of situations: 1) independent vehicle;2) between vehicle, existence is blocked;3) stick together at a distance of nearer two cars after binary image being corroded expansive working.Here, latter two situation is referred to as vehicle adhesion.Vehicle adhesion has a strong impact on vehicle tracking and vehicle Flow Detection, it would be highly desirable to a kind of reliable, practical Method of Vehicle Segmentation.At present, the method for adhesion vehicles segmentation includes:
1, the partitioning algorithm of feature based:
The characteristic point utilizing vehicle sets up characteristic model, it is possible to effectively solve adhesion problems, but, this type of existing algorithm operation quantity is big, and the impact being easily subject in complex scene uncorrelated object.
2, based on the partitioning algorithm of inference pattern:
Solve adhesion problems according to prioris such as the position of vehicle, speed, the directions of motion, this method many times can both dividing vehicle effectively, but be unsuitable for the outdoor traffic scene that circumstance complication is changeable.
3, the partitioning algorithm analyzed based on concavity:
Directly searching out degree of depth maximum point as preliminary election cut-point, in cell, corn etc. separate widely, but the method that they find cut-points is excessively complicated for this method.
Summary of the invention
For solving above-mentioned existing shortcoming, present invention is primarily targeted at provide a kind of practicality for splitting the method that there is adhesion vehicle in monitor video, algorithm is simple, and segmentation effect is reasonable.
For reaching above-described purpose, a kind of of the present invention adopts the following technical scheme that for splitting the method that there is adhesion vehicle in monitor video:
A kind of for splitting the method that there is adhesion vehicle in monitor video, it is characterised in that to comprise the steps:
(1) vehicle region detection: obtain foreground image by Gaussian Background modeling from video image, carries out cavity and fills and utilize burn into expansive working to eliminate noise, then obtain vehicle region by the method for four connected region labelling the foreground image obtained;
(2) adhesion vehicle region judges: occupied the area ratio of its boundary rectangle by vehicle region, and the length-width ratio of its boundary rectangle judges whether adhesion;
(3) vehicle region of adhesion is extracted skeleton with " baked wheaten cake method ", if adhesion performs step (7);
(4) detect skeleton angle point and cluster: the skeleton extracted is carried out Shi-Tomasi Corner Detection, and with K means clustering method the angle point classification detected;
(5) cut-off rule is determined: according to different adhesion situations, determine cut-off rule with diverse ways;
(6) segmentation adhesion vehicle;
(7) terminate.
In described step (1), the method for vehicle region detection is: utilize the method for Gaussian modeling to obtain video background, frame of video subtracting background just obtains prospect, the prospect obtained carries out cavity filling, corrosion expands pretreatment, each connected domain is marked by the method for four connected region labelling again after pretreatment, removing that area is less or the less region of number of pixels, remaining connected region is vehicle region.
In described step (2), adhesion vehicle region determination methods is: the area ratio of definition vehicle regionThe length-width ratio of boundary rectangleWherein AcIt is the area of vehicle region, ArIt is the area of vehicle region boundary rectangle, AhIt is the height of vehicle region boundary rectangle, AwIt it is the width of vehicle region boundary rectangle.
In described step (2), adhesion vehicle region determination methods is: set an area ratio threshold value Aat, and small one and large one two length-width ratio threshold value Alt1、Alt2Judge, work as Aa>AatAnd Alt2<Al<Alt1Shi Weifei adhesion vehicle region, otherwise for adhesion vehicle region;Simultaneously for adhesion vehicle Al<Alt2For left and right adhesion vehicle region, Al>Alt1For front and back adhesion vehicle region.
Described area ratio refers to the ratio of vehicle region area and its boundary rectangle area, and length-width ratio refers to the length of boundary rectangle and wide ratio;As a rule, the shape of non-adhesion vehicle becomes convex, and the space between boundary rectangle is little, and namely duty is relatively larger;And the length-width ratio of boundary rectangle is in certain scope.
The method in described step (3), with " baked wheaten cake method ", the vehicle region of adhesion being extracted skeleton is: removes some points from original figure, but still keeps original shape;For bianry image, it is judged that whether point removes is that concrete basis for estimation is: 1) internal point can not be deleted using the situation of around eight points (eight connectivity) that are connected as foundation;2) isolated point can not be deleted;3) straight line end points can not be deleted;4) point that removing connected component increases can not be deleted;The eight connectivity point around any non-zero pixels point P, P in image, is designated as P respectively clockwise about P point0, P1..., P7, wherein P0It is positioned at the upper left side place of eight connectivity;According to pixel value, P0P1P2P3P4P5P6P7A binary sequence can be formed, binary sequence is converted to decimal scale and will draw a certain number of 0~255, above four bases for estimation can obtain a concordance list, this table is a length is the array of 256, this array only comprises 0 or 1, when in array, nth is 1, then deletes this point;It is 0 not delete;Each in image puts the array label in manipulative indexing the exterior and the interior face, according to array label corresponding 0,1 judge whether this point is deleted.
The method in described step (3), with " baked wheaten cake method ", the vehicle region of adhesion being extracted skeleton is: first determine the binary sequence that the eight neighborhood of boundary point is corresponding, change into decimal scale N, concordance list is searched according to N value, when in table, n-th number is 1 deletion, it is 0 not delete, then more next boundary point is judged, circulate with this, delete layer by layer, until there is no deletable boundary point.
The method determining cut-off rule in described step (5) is: for the vehicle of front and back adhesion, find the highest angle point of the minimum angle point in the position of a class located above and a following class position, midpoint at two points does a horizontal line, and this horizontal line is exactly the cut-off rule of a car;For the vehicle of left and right adhesion, with the perpendicular bisector of two class cluster centre lines as cut-off rule.
Principles of the invention is in that: " skeleton " of an image refer to the bone portion skeleton of image central authorities can retain image topology character again can interference factor in subtracted image, reduce the interference of noise on image.Angle point refers to the point on boundary curve in image with curvature maximum, and in other words, acute variation can occur the gray value near this point, is the marked feature of piece image.The angle point of detection adhesion vehicle region skeleton, it is possible to flocked together by the angle point in same car, then utilizes K average Corner clustering, is easy to obtain cut-off rule according to the angle point of cluster, and algorithm is simply effective.
Adopt the present invention of as above technical scheme, have the advantages that
Based in the vehicle tracking of video, most vehicles segmentation algorithms need before speed, vehicle adhesion the prioris such as situation, and this method does not need any priori, this method makes full use of the framework characteristic of adhesion vehicle simultaneously, can effectively suppress the impact of noise on image.
Accompanying drawing explanation
Fig. 1 is the vehicles segmentation flow chart of the present invention.
Fig. 2 is the detection vehicle region flow chart of the present invention.
The front and back adhesion vehicle region figure that Fig. 3 (a) is the present invention.
The left and right adhesion vehicle region figure that Fig. 3 (b) is the present invention.
The adhesion vehicle region skeleton drawing that Fig. 4 (a) is the present invention.
The adhesion vehicle region skeleton angle point figure that Fig. 4 (b) is the present invention.
Fig. 5 is the Corner clustering skeleton drawing of the present invention.
Fig. 6 is the front and back adhesion vehicle region segmentation figure of the present invention.
Fig. 7 is the left and right adhesion vehicle region segmentation figure of the present invention.
First group of experimental result (used time 11.10ms) figure that Fig. 8 (a) is the present invention.
Second group of experimental result (used time 10.41ms) figure that Fig. 8 (b) is the present invention.
The 3rd group of experimental result (used time 8.96ms) figure that Fig. 8 (c) is the present invention.
Detailed description of the invention
In order to further illustrate the present invention, it is further illustrated below in conjunction with accompanying drawing:
Adhesion vehicle in various situations has been tested by algorithm in this paper, and experiment is on VS2008 platform, completes based on OpenCV2.4.4 storehouse.
As it is shown in figure 1, the flow process of vehicles segmentation of the present invention includes: vehicle region detection, the judgement of adhesion vehicle region, adhesion vehicle region skeletal extraction, detection skeleton angle point also carry out clustering, determining the vehicle of cut-off rule, segmentation adhesion.Followed by the explanation to each flow process.
Step 1, detection vehicle region:
As shown in Figure 2, detection vehicle region flow process in the present invention is: obtain video background by the method for Gaussian modeling, frame of video subtracting background just obtains prospect, the prospect obtained is carried out the pretreatment such as cavity filling, corrosion expansion, each connected domain is marked by the method for 4 connected component labelings again after pretreatment, removing that area is less or the less region of number of pixels, remaining connected region is vehicle region.
Step 2, judge vehicle whether adhesion:
The dutycycle of definition vehicle region
The length-width ratio of boundary rectangle
Wherein AcIt is the area of vehicle region, ArIt is the area of vehicle region boundary rectangle, AhIt is the height of vehicle region boundary rectangle, AwIt it is the width of vehicle region boundary rectangle.
For non-adhesion vehicle, AaVery big, then less for adhesion vehicle, the A of non-adhesion vehicle simultaneouslylSize, in certain scope, is set an area ratio threshold value A by experienceat, and small one and large one two length-width ratio threshold value Alt1、Alt2Judge, work as Aa>AatAnd Alt2<Al<Alt1Shi Weifei adhesion vehicle region, otherwise for adhesion vehicle region.Simultaneously for adhesion vehicle Al<Alt2For left and right adhesion vehicle region, as shown in Fig. 3 (b);Al>Alt1For front and back adhesion vehicle region, as shown in Fig. 3 (a).
Step 3, extraction adhesion vehicle region skeleton:
Having a lot of method extracting skeleton, wherein " baked wheaten cake method " is a kind of practical algorithm extracting bianry image target area.Being lighted everywhere by object boundary, the forward position of fire is at the uniform velocity to spread to target internal, and the fray-out of flame when forward position is intersected, the set of fray-out of flame point just constitutes axis, just defines the skeleton of target image simultaneously.The specific algorithm of " baked wheaten cake method " is as follows:
For bianry image, it is judged that whether point removes is that concrete basis for estimation is: 1) internal point can not be deleted using the situation of around 8 points (eight connectivity) that are connected as foundation;2) isolated point can not be deleted;3) straight line end points can not be deleted;4) point that removing connected component increases can not be deleted.The eight connectivity point around any non-zero pixels point P, P in image, is designated as P respectively clockwise about P point0, P1..., P7, wherein P0It is positioned at the upper left side place of eight connectivity.According to pixel value, P0P1P2P3P4P5P6P7Forming a binary sequence, binary sequence is converted to decimal scale and will draw a certain number of 0~255, by above four bases for estimation, we can obtain a concordance list, concordance list be one long be 256 array, this array only comprises 0 or 1.The array label in each the some eight connectivity situation manipulative indexing the exterior and the interior face around in image, according to array label corresponding 0,1 judge whether this point is deleted, 0 represents and does not delete, and 1 represents deletion.
" baked wheaten cake method " is that boundary point is processed, it is first determined the binary sequence that the eight neighborhood of boundary point is corresponding, changes into decimal scale N, concordance list is searched according to N value, when n-th number is 1 deletion in table, it is 0 and does not delete, then more next boundary point is judged, circulate with this, delete layer by layer, until there is no deletable boundary point, if Fig. 4 (a) is adhesion vehicle region skeleton drawing, wherein white portion is skeleton, and black part is divided into background.
Step 4, detection skeleton angle point:
Angle point refers to the point on boundary curve in image with curvature maximum, and in other words, acute variation can occur the gray value near this point, is the marked feature of piece image.It is single pixel framework due to what step 3 obtained, in order to highlight the flatness at non-key some place, we carry out Shi-Tomasi Corner Detection again the skeleton obtained after excessive erosion is expanded, Shi-Tomasi Corner Detection is the improvement of Harris algorithm, in a lot of situations, it is possible to obtain effect more better than use Harris algorithm.If Fig. 4 (b) is adhesion vehicle region skeleton angle point figure, wherein white portion is skeleton, and the round dot above skeleton is the angle point detected.
Each point on image first derivative both horizontally and vertically is the basis of algorithm, for any point I in image (u, v), its correspondence first derivative horizontally and vertically is:
Make A (u, v)=Ix 2(u,v)
B (u, v)=Iy 2(u,v)
C (u, v)=Ix(u,v)Iy(u,v)
These values are just called partial structurtes entry of a matrix element:
Then A, (u, v), (u, v), (u v) uses Gaussian filter H to C to B againGSmooth and obtain:
Due to matrixIt is be poised for battle matrix, so diagonalizable is:
Wherein λ1、λ2For matrixTwo characteristic roots, it is defined as:
In smooth image-regionThus λ1=λ2=0.Shi and Tomasi thinks, if one less in two eigenvalues more than minimum threshold, then can obtain angle point.It is illustrated in figure 5 adhesion vehicle region angle point figure.
Step 5, skeleton angle point cluster:
Kmeans clustering algorithm describes:
(1) 2 initial cluster centers are arbitrarily selected;
(2) calculate the distance between each object and two cluster centres, each object is distributed to the class apart from its nearest cluster centre place;
(3) class object and class center range difference is utilized to divide minimum principle to readjust class center;
(4) for 2 cluster centres of gained, after the iterative method renewal of (2), (3), value remains unchanged, then iteration terminates, and otherwise continues iteration.
The sharpest edges of this algorithm are in that succinct and quick.The selection that it is critical only that range formula of algorithm, the present invention selects Euclidean distance.As it is shown in figure 5, the round dot angle point in Fig. 4 (b) is divided into Fig. 5 orbicular spot and triangle two class.
Step 6, determine cut-off rule:
In step 2 adhesion vehicle region judges, may determine that adhesion region belongs to fore-aft vehicle adhesion or left and right vehicle adhesion according to boundary rectangle length-width ratio, according to both adhesion forms, determine cut-off rule with diverse ways respectively.
Adhesion segmentation before and after 6-1:
By step 5, we can obtain the skeleton drawing through Corner clustering, as it is shown in figure 5, it appeared that two class angle points separate up and down, and at the join domain of fore-aft vehicle, almost without angle point.And just can two classes separately with a horizontal line, then we look for the nethermost angle point of a class located above and a following uppermost angle point of class, midpoint at two points does a horizontal line, this horizontal line is exactly our cut-off rule, if Fig. 6 is the schematic diagram of adhesion vehicles segmentation before and after the present invention.
About 6-2 adhesion is split:
By step 5, we can obtain the cluster centre of two class angle points, and due to human vision problem, the cut-off rule of left and right adhesion is not simple vertical segment, and oblique line is in the majority.So for this kind of situation, we with the perpendicular bisector of two class cluster centre lines as cut-off rule.Such as Fig. 7, for adhesion vehicles segmentation schematic diagram in left and right of the present invention.
Fig. 8 (a), Fig. 8 (b), Fig. 8 (c) are part measured result figure of the present invention, and wherein (ai) (i is 1,2,3 herein) represents each original vehicle figure;(bi) representing the vehicle foreground of extraction for white portion in prospect adhesion figure, figure, black part submeter shows background;(ci) represent by after skeletal extraction, Corner Detection and cluster through the vehicle foreground of over-segmentation, the cut-off rule of white portion is the cut-off rule determined according to vehicle adhesion situation;(di) for prospect skeleton drawing, white portion is that the skeleton of vehicle region, round dot and triangle represent a class angle point respectively;(ei) for the vehicle figure after segmentation, the vehicle split with square frame frame place, only has a car respectively in each square frame.
Above by description of listed embodiment, elaborate the basic ideas and basic principles of the present invention.But the present invention is not limited to above-mentioned listed embodiment, every equivalent variations, improvement made based on technical scheme and the behavior such as deliberately deteriorate, protection scope of the present invention all should be belonged to.
Claims (8)
1. one kind is used for splitting the method that there is adhesion vehicle in monitor video, it is characterised in that comprise the steps:
(1) vehicle region detection: obtain foreground image by Gaussian Background modeling from video image, carries out cavity and fills and utilize burn into expansive working to eliminate noise, then obtain vehicle region by the method for four connected region labelling the foreground image obtained;
(2) adhesion vehicle region judges: occupied the area ratio of its boundary rectangle by vehicle region, and the length-width ratio of its boundary rectangle judges whether adhesion;
(3) vehicle region of adhesion is extracted skeleton with " baked wheaten cake method ", if adhesion performs step (7);
(4) detect skeleton angle point and cluster: the skeleton extracted is carried out Shi-Tomasi Corner Detection, and with K means clustering method the angle point classification detected;
(5) cut-off rule is determined: according to different adhesion situations, determine cut-off rule with diverse ways;
(6) segmentation adhesion vehicle;
(7) terminate.
2. according to claim 1 a kind of for splitting the method that there is adhesion vehicle in monitor video, it is characterized in that, in described step (1), the method for vehicle region detection is: utilize the method for Gaussian modeling to obtain video background, frame of video subtracting background just obtains prospect, the prospect obtained carries out cavity filling, corrosion expands pretreatment, each connected domain is marked by the method for four connected region labelling again after pretreatment, removing that area is less or the less region of number of pixels, remaining connected region is vehicle region.
3. according to claim 1 a kind of for splitting the method that there is adhesion vehicle in monitor video, it is characterised in that in described step (2), adhesion vehicle region determination methods is: the area ratio of definition vehicle region;The length-width ratio of boundary rectangle;It is wherein the area of vehicle region, is the area of vehicle region boundary rectangle, be the height of vehicle region boundary rectangle, be the width of vehicle region boundary rectangle.
4. a kind of for splitting the method that there is adhesion vehicle in monitor video according to claim 1 or 3, it is characterized in that, in described step (2), adhesion vehicle region determination methods is: set an area ratio threshold value, and small one and large one two length-width ratio threshold values, judges, when>And < < time be non-adhesion vehicle region, otherwise for adhesion vehicle region;It is front and back adhesion vehicle region simultaneously for adhesion vehicle<for left and right adhesion vehicle region,>.
5. a kind of for splitting the method that there is adhesion vehicle in monitor video according to claim 1 or 3, it is characterised in that described area ratio refers to the ratio of vehicle region area and its boundary rectangle area, and length-width ratio refers to the length of boundary rectangle and wide ratio;As a rule, the shape of non-adhesion vehicle becomes convex, and the space between boundary rectangle is little, and namely duty is relatively larger;And the length-width ratio of boundary rectangle is in certain scope.
6. according to claim 1 a kind of for splitting the method that there is adhesion vehicle in monitor video, it is characterized in that, the method in described step (3), with " baked wheaten cake method ", the vehicle region of adhesion being extracted skeleton is: removes some points from original figure, but still keeps original shape;For bianry image, it is judged that whether point removes is that concrete basis for estimation is: 1) internal point can not be deleted using the situation of around eight points (eight connectivity) that are connected as foundation;2) isolated point can not be deleted;3) straight line end points can not be deleted;4) point that removing connected component increases can not be deleted;The eight connectivity point around any non-zero pixels point P, P in image, is designated as P respectively clockwise about P point0, P1..., P7, wherein P0It is positioned at the upper left side place of eight connectivity;According to pixel value, P0P1P2P3P4P5P6P7A binary sequence can be formed, binary sequence is converted to decimal scale and will draw a certain number of 0 ~ 255, above four bases for estimation can obtain a concordance list, this table is a length is the array of 256, this array only comprises 0 or 1, when in array, nth is 1, then deletes this point;It is 0 not delete;Each in image puts the array label in manipulative indexing the exterior and the interior face, according to array label corresponding 0,1 judge whether this point is deleted.
7. a kind of for splitting the method that there is adhesion vehicle in monitor video according to claim 1 or 6, it is characterized in that, the method in described step (3), with " baked wheaten cake method ", the vehicle region of adhesion being extracted skeleton is: first determine the binary sequence that the eight neighborhood of boundary point is corresponding, change into decimal scale N, concordance list is searched according to N value, when in table, n-th number is 1 deletion, it is 0 not delete, then more next boundary point is judged, circulate with this, delete layer by layer, until there is no deletable boundary point.
8. according to claim 1 a kind of for splitting the method that there is adhesion vehicle in monitor video, it is characterized in that, the method determining cut-off rule in described step (5) is: for the vehicle of front and back adhesion, find the highest angle point of the minimum angle point in the position of a class located above and a following class position, midpoint at two points does a horizontal line, and this horizontal line is exactly the cut-off rule of a car;For the vehicle of left and right adhesion, with the perpendicular bisector of two class cluster centre lines as cut-off rule.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510109679.2A CN105809167B (en) | 2015-03-13 | 2015-03-13 | A method of for dividing, there are adhesion vehicles in monitor video |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510109679.2A CN105809167B (en) | 2015-03-13 | 2015-03-13 | A method of for dividing, there are adhesion vehicles in monitor video |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105809167A true CN105809167A (en) | 2016-07-27 |
CN105809167B CN105809167B (en) | 2019-06-18 |
Family
ID=56465587
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510109679.2A Active CN105809167B (en) | 2015-03-13 | 2015-03-13 | A method of for dividing, there are adhesion vehicles in monitor video |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105809167B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108694402A (en) * | 2017-04-07 | 2018-10-23 | 富士通株式会社 | String segmentation device and method, character string identification device and method |
CN109816086A (en) * | 2017-11-20 | 2019-05-28 | 富士通株式会社 | Counting device, method and the electronic equipment of mobile object |
CN110348363A (en) * | 2019-07-05 | 2019-10-18 | 西安邮电大学 | The vehicle tracking algorithm for eliminating similar vehicle interference is merged based on multiframe angle information |
CN113538500A (en) * | 2021-09-10 | 2021-10-22 | 科大讯飞(苏州)科技有限公司 | Image segmentation method and device, electronic equipment and storage medium |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101251927A (en) * | 2008-04-01 | 2008-08-27 | 东南大学 | Vehicle detecting and tracing method based on video technique |
EP2665018A1 (en) * | 2012-05-14 | 2013-11-20 | Thomson Licensing | Object identification in images or image sequences |
-
2015
- 2015-03-13 CN CN201510109679.2A patent/CN105809167B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101251927A (en) * | 2008-04-01 | 2008-08-27 | 东南大学 | Vehicle detecting and tracing method based on video technique |
EP2665018A1 (en) * | 2012-05-14 | 2013-11-20 | Thomson Licensing | Object identification in images or image sequences |
Non-Patent Citations (1)
Title |
---|
刘诗慧: "视频中粘连车辆的分割与跟踪技术研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108694402A (en) * | 2017-04-07 | 2018-10-23 | 富士通株式会社 | String segmentation device and method, character string identification device and method |
CN108694402B (en) * | 2017-04-07 | 2021-10-01 | 富士通株式会社 | Character string segmentation device and method, character string recognition device and method |
CN109816086A (en) * | 2017-11-20 | 2019-05-28 | 富士通株式会社 | Counting device, method and the electronic equipment of mobile object |
CN109816086B (en) * | 2017-11-20 | 2023-05-23 | 富士通株式会社 | Counting device and method for moving object and electronic equipment |
CN110348363A (en) * | 2019-07-05 | 2019-10-18 | 西安邮电大学 | The vehicle tracking algorithm for eliminating similar vehicle interference is merged based on multiframe angle information |
CN110348363B (en) * | 2019-07-05 | 2021-06-15 | 西安邮电大学 | Vehicle tracking method for eliminating similar vehicle interference based on multi-frame angle information fusion |
CN113538500A (en) * | 2021-09-10 | 2021-10-22 | 科大讯飞(苏州)科技有限公司 | Image segmentation method and device, electronic equipment and storage medium |
CN113538500B (en) * | 2021-09-10 | 2022-03-15 | 科大讯飞(苏州)科技有限公司 | Image segmentation method and device, electronic equipment and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN105809167B (en) | 2019-06-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Tan et al. | Color model-based real-time learning for road following | |
CN107993245B (en) | Aerospace background multi-target detection and tracking method | |
CN108537239B (en) | Method for detecting image saliency target | |
CN103106667A (en) | Motion target tracing method towards shielding and scene change | |
CN108985169A (en) | Across the door operation detection method in shop based on deep learning target detection and dynamic background modeling | |
CN105809167A (en) | Method for parting vehicles sticking together in monitoring video | |
CN103871076A (en) | Moving object extraction method based on optical flow method and superpixel division | |
CN106204594A (en) | A kind of direction detection method of dispersivity moving object based on video image | |
CN107862702A (en) | A kind of conspicuousness detection method of combination boundary connected and local contrast | |
CN110309765B (en) | High-efficiency detection method for video moving target | |
CN105321189A (en) | Complex environment target tracking method based on continuous adaptive mean shift multi-feature fusion | |
CN103578083A (en) | Single image defogging method based on joint mean shift | |
CN102592128A (en) | Method and device for detecting and processing dynamic image and display terminal | |
CN104732559A (en) | Multi-target detecting and tracking method based on RGB-D data | |
Mayr et al. | Self-supervised learning of the drivable area for autonomous vehicles | |
CN103559725A (en) | Wireless sensor node optimization selection method orientated at visual tracking | |
CN110443142B (en) | Deep learning vehicle counting method based on road surface extraction and segmentation | |
CN104484865A (en) | Method for removing raindrops in video image | |
CN106056545A (en) | Image rain removing method and image rain removing system | |
CN107871315B (en) | Video image motion detection method and device | |
CN103366158A (en) | Three dimensional structure and color model-based monocular visual road face detection method | |
CN103218828B (en) | A kind of many cells interaction tracker | |
Ilehag et al. | Classification and representation of commonly used roofing material using multisensorial aerial data | |
CN110059544B (en) | Pedestrian detection method and system based on road scene | |
Bhaskar et al. | Enhanced and effective parallel optical flow method for vehicle detection and tracking |
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 |