CN102375983B - Based on the image segmentation processing method of area matching optimization K-means clustering algorithm - Google Patents
Based on the image segmentation processing method of area matching optimization K-means clustering algorithm Download PDFInfo
- Publication number
- CN102375983B CN102375983B CN201010254746.7A CN201010254746A CN102375983B CN 102375983 B CN102375983 B CN 102375983B CN 201010254746 A CN201010254746 A CN 201010254746A CN 102375983 B CN102375983 B CN 102375983B
- Authority
- CN
- China
- Prior art keywords
- segmentation
- cluster
- area
- image
- clustering algorithm
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Image Analysis (AREA)
Abstract
The present invention relates to the image segmentation processing method based on area matching optimization K-means clustering algorithm in the detection of a kind of video, first it extract the vehicle characteristics point of front and back two field picture, then the area overlapping cases of contrast front and back two frame vehicles, extract characteristic point position and residue character point position in area overlapping region, ask for two stack features point averages respectively as 2 class initial cluster center points to be split, then the segmentation of K-average is carried out, according to the classification situation of unique point in the cluster result correction area overlapping region exported, judge that whether the vehicle after cluster is reasonable simultaneously, as unreasonable, then to cluster result again cluster and Statistical Clustering Analysis center again, until terminate cluster segmentation after finding rational vehicle, returning tracking result.The method based on area matched optimization, and adopts fixing cluster number to split, and the vehicle target obtained after the segmentation of K-average no longer needs to enter the matching treatment of next round, makes processing speed faster, has saved the time.
Description
Technical field
The present invention relates to video detection technology field, particularly video image target tracking technique field, specifically refer to the image segmentation processing method based on area matching optimization K-means clustering algorithm.
Background technology
First video frequency vehicle Tracking Recognition technology will be partitioned into moving target in video image, then could tracking target.When traffic is very crowded, target vehicle may overlap with other vehicle, and now, traditional partitioning algorithm is difficult to be partitioned into the connected region comprising single unit vehicle.In order to overcome this problem, adopting the method tracking target vehicle following the tracks of vehicle characteristics point at present more and more, now just needing to split, to improve tracking efficiency many vehicle characteristics point of adhesion.
Traditional K-means clustering algorithm does not choose initial cluster center by random approach, the difference of selected point, and cluster result may be just different, and such dependence just causes the instability of cluster result, and is easily absorbed in local optimum but not global optimum's cluster result; And this clustering algorithm to noise spot and isolated point very sensitive; Cluster result depends on the setting of initial value, but selected often will the experiment through many times of k value (cluster number) just can find optimum value.Although there is a lot of innovatory algorithm at present, how correctly to determine that initial cluster center and cluster number remain the technical matters needed badly and overcome.
Summary of the invention
The object of the invention is to overcome above-mentioned shortcoming of the prior art, provide a kind of and introduce area matching algorithm, fixing cluster number, based on the image segmentation processing method of area matching optimization K-means clustering algorithm in processing speed video detection faster.
In order to realize above-mentioned object, the image segmentation processing method based on area matching optimization K-means clustering algorithm during video of the present invention detects comprises the following steps:
(0) after extracting current frame motion target image, by mating the target image finding movement destination image in former frame with described to match;
(1) present frame and former frame motion characteristics point is extracted respectively;
(2) carry out area matched to present frame and former frame, obtain the overlapping area region that current frame motion target image is overlapping with former frame movement destination image;
(3) the overlapping area region of present frame and the unique point average of non-overlapped surface area is calculated respectively, as K-average initial cluster center;
(4) carry out the segmentation of K-mean cluster, and revise segmentation result;
(5) judge that whether segmentation is successful, if success, then enter step (8), if unsuccessful, then enter step (6);
(6) whether unanimously with a front segmentation result compare this segmentation result, if unanimously, then enter step (8), if inconsistent, then enter step (7);
(7) according to the result adjustment initial cluster center after segmentation, and step (4) is returned;
(8) Output rusults, method ends.
Should based in the image segmentation processing method of area matching optimization K-means clustering algorithm, described K-mean cluster is divided into the segmentation of 2-mean cluster.
Should based in the image segmentation processing method of area matching optimization K-means clustering algorithm, described unique point comprises area features point, position feature point and Feature Points.
Should based in the image segmentation processing method of area matching optimization K-means clustering algorithm, described step (4) is specially: carry out the segmentation of K-mean cluster, and according to cluster result, revises the classification situation of eigenwert in overlapping area region.
Should based in the image segmentation processing method of area matching optimization K-means clustering algorithm, whether judging in described step (5) is split successful, is specially: judge that whether the area of the target image after cluster is reasonable according to the area of previous frame target image.
Should based in the image segmentation processing method of area matching optimization K-means clustering algorithm, described adjusts initial cluster center according to the result after segmentation, be specially: (7) are according to the result of segmentation, be limited with the cluster feature point that there is target image, readjust initial cluster center.
Should based in the image segmentation processing method of area matching optimization K-means clustering algorithm, described movement destination image is vehicle image.
Have employed the image segmentation processing method based on area matching optimization K-means clustering algorithm in video of the present invention detection, its vehicle characteristics point first extracting front and back two field picture (comprises area, position, shape), because camera shutter speed is enough fast, so certainly there is overlap in front and back two frame vehicle area.The area overlapping cases of two frame vehicles before and after contrast, tentatively can obtain the position of following the tracks of vehicle headstock or the tailstock.If a rear frame vehicle and other vehicle exist adhesion, extract characteristic point position in area overlapping region, ask for these unique point averages, extract residue character point position simultaneously and ask for its average.Two averages now obtained are respectively as 2 class initial cluster center points to be split.Then the segmentation of K-average is carried out, according to the classification situation of unique point in the above-mentioned area overlapping region of cluster result correction exported, simultaneously judge that whether the vehicle after now cluster is reasonable, if unreasonable, then to cluster result again cluster according to previous frame vehicle area.Now only there is that block cluster feature group of vehicle to be tracked in segmentation, simultaneously according to above method again Statistical Clustering Analysis center.Until terminate cluster segmentation after finding rational vehicle, returning tracking result.If cluster result does not have difference after segmentation repeatedly, and does not find suitable cluster result, then enter next step matching treatment.The method is based on area matched optimization, and adopt fixing cluster number to split, the vehicle target obtained after the segmentation of K-average no longer needs to enter the matching treatment of next round, makes processing speed faster, has saved the processing time that overall video image target is followed the tracks of.
Accompanying drawing explanation
Fig. 1 is the process flow diagram based on the image segmentation processing method of area matching optimization K-means clustering algorithm during video of the present invention detects.
Fig. 2 is the process flow diagram of area matching optimization K-means clustering algorithm of the present invention.
Fig. 3 is the schematic diagram of feature extraction in area matching optimization K-means clustering algorithm of the present invention.
Fig. 4 is that in area matching optimization K-means clustering algorithm of the present invention, middle cluster centre extracts schematic diagram.
Embodiment
In order to more clearly understand technology contents of the present invention, describe in detail especially exemplified by following examples.
Referring to shown in Fig. 1, is the process flow diagram based on a kind of embodiment of the image segmentation processing method of area matching optimization K-means clustering algorithm in video detection of the present invention.
In this embodiment, described method comprises the following steps:
(0) after extracting current frame motion target image, by mating the target image finding movement destination image in former frame with described to match;
(1) present frame and former frame motion characteristics point is extracted respectively;
(2) carry out area matched to present frame and former frame, obtain the overlapping area region that current frame motion target image is overlapping with former frame movement destination image;
(3) the overlapping area region of present frame and the unique point average of non-overlapped surface area is calculated respectively, as K-average initial cluster center;
(4) carry out the segmentation of K-mean cluster, and revise segmentation result;
(5) judge that whether segmentation is successful, if success, then enter step (8), if unsuccessful, then enter step (6);
(6) whether unanimously with a front segmentation result compare this segmentation result, if unanimously, then enter step (8), if inconsistent, then enter step (7);
(7) according to the result adjustment initial cluster center after segmentation, and step (4) is returned;
(8) Output rusults, method ends.
In this embodiment, described movement destination image is vehicle image, and described K-mean cluster is divided into the segmentation of 2-mean cluster, and described unique point comprises area features point, position feature point and Feature Points.
One of the present invention preferred embodiment in, the step (4) based on the image segmentation processing method of area matching optimization K-means clustering algorithm during described video detects is specially:
Carry out the segmentation of K-mean cluster, and according to cluster result, revise the classification situation of eigenwert in overlapping area region.
In another preferred embodiment of the present invention, whether successful split based on judging in the step (5) of the image segmentation processing method of area matching optimization K-means clustering algorithm during described video detects, be specially:
Judge that whether the area of the target image after cluster is reasonable according to the area of previous frame target image.
Described adjusts initial cluster center according to the result after segmentation, is specially:
(7) according to the result of segmentation, be limited with the cluster feature point that there is target image, readjust initial cluster center.
In practical application of the present invention, the flow process of the K-means clustering algorithm of the improvement adopted as shown in Figure 2, comprises the following steps:
(1), after extracting current frame motion target, find by coupling the former frame vehicle matched with it;
(2) as shown in Figure 3, present frame and former frame motion characteristics point is extracted respectively;
(3) as shown in Figure 4, carry out area matched to two vehicles, obtain overlapping area region;
(4) as shown in Figure 4, the unique point average of the present frame of overlapping area region and non-overlapped surface area is calculated respectively, as K-average initial cluster center;
(5) carry out the segmentation of 2-mean cluster, revise segmentation result, judge that whether segmentation is successful; If unsuccessful, according to the result adjustment initial cluster center after segmentation,
(6) repeat step 5), until split successfully or split restrain;
(7) Output rusults.
Generally, need many vehicles phenomenon that sticks together in the region of following the tracks of relatively less, and vehicle characteristics point is few, so the operand of this algorithm little, processing speed is faster; If cluster result does not have difference after segmentation repeatedly, and does not find suitable cluster result, then enter next step matching treatment, in the case, repeat cutting procedure although exist, namely also can ensure real-time follow-up.
K-average dividing method after improvement, it is based on area matched optimization, and adopt fixing cluster number to split, the vehicle target obtained after the segmentation of K-average no longer needs to enter the matching treatment of next round, make processing speed faster, save the processing time that overall video image target is followed the tracks of.
In this description, the present invention is described with reference to its specific embodiment.But, still can make various amendment and conversion obviously and not deviate from the spirit and scope of the present invention.Therefore, instructions and accompanying drawing are regarded in an illustrative, rather than a restrictive.
Claims (7)
1. video detect in based on the image segmentation processing method of area matching optimization K-means clustering algorithm, it is characterized in that, described method comprises the following steps:
(0) after extracting current frame motion target image, by mating the target image finding movement destination image in former frame with described to match;
(1) present frame and former frame motion characteristics point is extracted respectively;
(2) carry out area matched to present frame and former frame, obtain the overlapping area region that current frame motion target image is overlapping with former frame movement destination image;
(3) the overlapping area region of present frame and the unique point average of non-overlapped surface area is calculated respectively, as K-average initial cluster center;
(4) carry out the segmentation of K-mean cluster, and revise segmentation result;
(5) judge that whether segmentation is successful, if success, then enter step (8), if unsuccessful, then enter step (6);
(6) whether unanimously with a front segmentation result compare this segmentation result, if unanimously, then enter step (8), if inconsistent, then enter step (7);
(7) according to the result adjustment initial cluster center after segmentation, and step (4) is returned;
(8) Output rusults, method ends.
2. video according to claim 1 detect in based on the image segmentation processing method of area matching optimization K-means clustering algorithm, it is characterized in that, described K-mean cluster is divided into the segmentation of 2-mean cluster.
3. video according to claim 1 detect in based on the image segmentation processing method of area matching optimization K-means clustering algorithm, it is characterized in that, described unique point comprises area features point, position feature point and Feature Points.
4. video according to claim 1 detect in based on the image segmentation processing method of area matching optimization K-means clustering algorithm, it is characterized in that, described step (4) is specially:
Carry out the segmentation of K-mean cluster, and according to cluster result, revise the classification situation of unique point average in overlapping area region.
5. video according to claim 1 detect in based on the image segmentation processing method of area matching optimization K-means clustering algorithm, it is characterized in that, in described step (5), judging segmentation whether success, be specially:
Judge that whether the area of the target image after cluster is reasonable according to the area of former frame target image.
6. video according to claim 1 detect in based on the image segmentation processing method of area matching optimization K-means clustering algorithm, it is characterized in that, described according to the result adjustment initial cluster center after segmentation, be specially:
(7) according to the result of segmentation, be limited with the cluster feature point that there is target image, readjust initial cluster center.
7. video according to any one of claim 1 to 6 detect in based on the image segmentation processing method of area matching optimization K-means clustering algorithm, it is characterized in that, described movement destination image is vehicle image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201010254746.7A CN102375983B (en) | 2010-08-17 | 2010-08-17 | Based on the image segmentation processing method of area matching optimization K-means clustering algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201010254746.7A CN102375983B (en) | 2010-08-17 | 2010-08-17 | Based on the image segmentation processing method of area matching optimization K-means clustering algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102375983A CN102375983A (en) | 2012-03-14 |
| CN102375983B true CN102375983B (en) | 2015-11-25 |
Family
ID=45794563
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201010254746.7A Active CN102375983B (en) | 2010-08-17 | 2010-08-17 | Based on the image segmentation processing method of area matching optimization K-means clustering algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102375983B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105930833B (en) * | 2016-05-19 | 2019-01-22 | 重庆邮电大学 | A kind of vehicle tracking and dividing method based on video monitoring |
| CN106570877B (en) * | 2016-10-27 | 2017-12-05 | 西安科技大学 | Based on coal-winning machine virtual prototype the coal-winning machine pose alignment system and method registering with true picture |
| CN109800684B (en) * | 2018-12-29 | 2022-06-21 | 上海依图网络科技有限公司 | Method and device for determining object in video |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1575014A2 (en) * | 2001-06-21 | 2005-09-14 | Fujitsu Limited | Method and apparatus for processing pictures of vehicles |
| CN101159855A (en) * | 2007-11-14 | 2008-04-09 | 南京优科漫科技有限公司 | Characteristic point analysis based multi-target separation predicting method |
-
2010
- 2010-08-17 CN CN201010254746.7A patent/CN102375983B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1575014A2 (en) * | 2001-06-21 | 2005-09-14 | Fujitsu Limited | Method and apparatus for processing pictures of vehicles |
| CN101159855A (en) * | 2007-11-14 | 2008-04-09 | 南京优科漫科技有限公司 | Characteristic point analysis based multi-target separation predicting method |
Non-Patent Citations (1)
| Title |
|---|
| 鲁棒的实时多车辆检测与跟踪系统设计;陈功，杨奎元，周荷琴，严捷丰;《信号处理》;20090430;第25卷(第4期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102375983A (en) | 2012-03-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108304808B (en) | Monitoring video object detection method based on temporal-spatial information and deep network | |
| CN101448100B (en) | Method for extracting video captions quickly and accurately | |
| US20180089505A1 (en) | System and method for deep network fusion for fast and robust object detection | |
| Wang et al. | Object detection combining recognition and segmentation | |
| CN107578091B (en) | Pedestrian and vehicle real-time detection method based on lightweight deep network | |
| Enzweiler et al. | Efficient stixel-based object recognition | |
| CN102375983B (en) | Based on the image segmentation processing method of area matching optimization K-means clustering algorithm | |
| Lu et al. | A novel approach for video text detection and recognition based on a corner response feature map and transferred deep convolutional neural network | |
| CN111539907B (en) | Image processing method and device for target detection | |
| CN102902957A (en) | Video-stream-based automatic license plate recognition method | |
| Cao et al. | Multi-view frustum pointnet for object detection in autonomous driving | |
| CN103927763B (en) | A kind of image sequence multiple target tracking track identification processing method | |
| WO2016059643A1 (en) | System and method for pedestrian detection | |
| CN113901897A (en) | Parking lot vehicle detection method based on DARFNet model | |
| Liu et al. | Adaptive pattern-parameter matching for robust pedestrian detection | |
| CN103886609A (en) | Vehicle tracking method based on particle filtering and LBP features | |
| CN109255052B (en) | Three-stage vehicle retrieval method based on multiple features | |
| Hsieh et al. | A shadow elimination method for vehicle analysis | |
| Behera et al. | A novel method for day time pedestrian detection | |
| CN102521841A (en) | Multi-target object tracking method | |
| Kim et al. | Skyline extraction using a multistage edge filtering | |
| CN106127235A (en) | A kind of vehicle query method and system based on target characteristic collision | |
| Arnell et al. | Fast object segmentation from a moving camera | |
| Libiao et al. | Semantic segmentation based on DeeplabV3+ with multiple fusions of low-level features | |
| CN116615666A (en) | Sequence processing for data sets with lost frames |
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 |