CN109102520A - The moving target detecting method combined based on fuzzy means clustering with Kalman filter tracking - Google Patents
The moving target detecting method combined based on fuzzy means clustering with Kalman filter tracking Download PDFInfo
- Publication number
- CN109102520A CN109102520A CN201810549503.2A CN201810549503A CN109102520A CN 109102520 A CN109102520 A CN 109102520A CN 201810549503 A CN201810549503 A CN 201810549503A CN 109102520 A CN109102520 A CN 109102520A
- Authority
- CN
- China
- Prior art keywords
- fuzzy
- color difference
- block
- kalman filter
- means clustering
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/277—Analysis of motion involving stochastic approaches, e.g. using Kalman filters
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/90—Determination of colour characteristics
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The present invention discloses a kind of moving target detecting method combined based on fuzzy means clustering with Kalman filter tracking, the main thought of this method is: first calculating the color difference histogram (CDH) of pixel local neighborhood, then fuzzy color difference histogram is obtained using Fuzzy c-means Clustering (FCM), and then background modeling is carried out, and similitude matching detection prospect is used, target tracking is further carried out using the Kalman filter based on Block- matching.The background subtraction mechanism based on fuzzy color difference histogram FCDH proposed replaces Pixel-level using patch grade method, and this method is focused on color difference, rather than in color size.By reducing the quantity of false error, FCDH can show excellent performance.
Description
Technical field
The invention belongs to multiple technical fields such as information science, computer vision, machine learning, pattern-recognition, especially
It is related to a kind of moving target detecting method combined based on fuzzy means clustering with Kalman filter tracking.
Background technique
Currently, monitoring video information is automatically processed with prediction in information science, computer vision, machine learning, mode
It is greatly paid close attention in the multiple fields such as identification.And the foreground target information in monitor video how effectively, is quickly extracted,
It is wherein extremely important and basic problem.This technology has been widely used in video object tracking, and urban transportation detection is long
When scene monitor, video actions capture, video compress etc. application in.
Traditional moving target detecting method includes frame difference method, background subtraction method and optical flow method.Frame difference method is continuous
The adjacent interframe of two or three in image sequence is extracted in image using time difference pixel-based and thresholding
Moving target.The advantages of frame difference method is that algorithm is simple, and speed is fast, is easy to hardware realization, and it is big can to well adapt to environmental change
The case where, but for moving slow target in image sequence, it is difficult to it effectively detects complete prospect, is easy inside movement entity
Generate larger " cavity ".Optical flow method is by establishing target motion vectors field, with change of the pixel in image sequence in time-domain
Change and consecutive frame between correlation find previous frame with corresponding relationship existing between present frame, to calculate adjacent
A kind of method of the motion information of object between frame.Optical flow method can also examine moving target in movement background environment
It surveys, but its is maximum the disadvantage is that calculation amount is excessive, it is difficult to accomplish real-time detection.
When carrying out moving object detection, not only target itself, but also be affected by other factors, such as change
Illumination, dynamic background, crowded scene and block.In the case where video camera is unstable, moving object segmentation problem becomes
It is more challenging because background be no longer it is static, background subtraction method can no longer be used.
Summary of the invention
In order to overcome the shortcomings of the prior art described above, a kind of poly- based on fuzzy mean it is an object of the invention to propose
The moving target detecting method that class is combined with Kalman filter tracking, the background based on fuzzy color difference histogram FCDH of proposition
Subduction mechanism replaces Pixel-level using patch grade method, and this method is focused on color difference, rather than in color size.
By reducing the quantity of false error, FCDH can show excellent performance.
In order to achieve the above object, the technical scheme adopted by the invention is that: filtered based on fuzzy means clustering and Kalman
Wave tracks the moving target detecting method combined, which is characterized in that described method includes following steps:
Step 1, the color difference histogram CDH of pixel local neighborhood is calculated;
Step 2, fuzzy color difference histogram is obtained using Fuzzy c-means Clustering FCM;
Step 3, background modeling is carried out, and uses similitude matching detection prospect;
Step 4, target tracking is carried out using the Kalman filter based on Block- matching.
Further, step 1 calculates the color difference of the local neighborhood of pixel using the color difference for corresponding to two color components
Histogram CDH, specifically includes;
1.1 calculating values of chromatism: all frames in video-frequency band are read in and are saved, are denoted as I respectively1, I2..., Im∈Nm×n,
Wherein, M is the totalframes in video-frequency band, and m, n are the size of every frame picture;There to be the cromogram of strength grade I (p, q, ch)
As being quantified as W grade, I ∈ 0,1,2 ... W-1 };(p, q) is position coordinates, and ch is Color Channel, and RGB image is converted
For CIE L*a*b* color model;
The color difference d in small part N × N neighborhood centered on position (p, q) is sought, is calculated using following formula:
Wherein, I (r, s, ch) indicates the intensity of pixel in neighborhood;
1.2 color difference histograms: it will be blurred from (1) calculated color difference d using Gauss member function are as follows:
σ indicates standard deviation;
Color difference histogram is finally given by the following formula:
Wherein, A × A is the regional area centered on the position (p, q), and the frequency of pixel is calculated with k.
Further, step 2 obtains fuzzy color difference histogram using Fuzzy c-means Clustering FCM, specifically includes:
FCM is by K-bin local histogram H={ h1, h2, h3..., hkBe classified as with liC cluster centered on position, often
A cluster distributes bin based on fuzzy membership, this is completed by minimizing the iteration of cost function;
Fuzzy color difference histogram uses subordinated-degree matrix u (c × K dimension) and CDH vector H (K × 1) building c Victoria C DH (FCDH)
Fuzzy vector v, as follows:
hc×1=uc×KHK×1 (5)。
Further, step 3 carries out background modeling, and uses similitude matching detection prospect, specifically includes;
3.1 background modelings: similarity measurement p is between modeling background frame F and present frame h using histogram measurement
The intersection of FCDH;
Wherein, c is the quantity for providing histogram case, and background B is made comparisons by same threshold value and obtained:
3.2 context updates: background FCDH uses [0,1] to update each pixel as Background learning rate, and t is time index,
It is as follows:
Further, step 4 carries out target tracking using the Kalman filter based on Block- matching, specifically includes:
4.1 can be changed block size (BVBS) BM from bottom to top: just starting to select small size block to detect more moving mass,
Then block size doubles in each dimension, a series of pieces of { Bi, i=1,2,3,4 } } in smaller size, they are shared
Identical larger size father block Sb, it is defined as follows:
Sb=B1∪B2∪B3∪B4, where BI, 1≤1≤4
1,2,3 and 4 father's block S is corresponded respectively tobUpper left, upper right, lower-left, bottom right sub-block;
B in step previousiDetecting the BM process with small moving mass can be repeated as desired for repeatedly, or until block
Until size reaches scheduled full-size, in this way, more false moving mass can be abandoned with biggish block size, keep simultaneously
The real motion details of smaller piece size;
4.2 times updated: known state xk-1, start iteration to NextState xkIt is predicted;
Wherein,It is prior estimate error covariance, because the doing exercises for object block in X and Y coordinates is complete phase
Mutually independent, Kalman filtering can apply X, Y coordinates respectively, and in each coordinate, the fast movement of a target can be to
Lower formula indicates:
xk,Δ k respectively indicates the displacement at k moment, speed, acceleration and the time difference, because not knowing control item,
B in equationkukIt is 0, the specific equation of the motion tracking of real-time update is as follows:
The determination of process noise covariance is generally relatively difficult, because the estimation procedure directly observed is not always feasible
, it, can be using acceleration as piecewise constant Wiener process in this work, i.e. acceleration continues in each period
It is remained unchanged in time, but different for each period, and is each zero-mean gaussian sequence,
Q is that the energy spectral density density of continuous time white noise enables P to reduce calculation amount0=cQk, c is one normal
Number;
4.3: measurement updaue: state xkZkIt can be indicated with a linear equation:
zk=Hkxk+vk (13)
zkIt is the measured value of m × 1 at k moment, HkIt is and zkThe observing matrix of relevant m × n, vkIt is the additional survey of m × 1
Noise vector is measured, the equation of specific measurement updaue is as follows:
Formula n × m matrix K is so that the smallest gain of posteriori error covariance.
Compared with prior art, the beneficial effects of the present invention are: the invention proposes a kind of fuzzy means clustering and karrs
The moving target detecting method that graceful filter tracking combines, compared with the conventional method, remarkable advantage is:
A) the color difference histogram calculation of local neighborhood can handle due to dynamic background (such as tap water, brandish trees) and
The problem.
B) fuzzy color difference histogram is obtained using Fuzzy c-means Clustering FCM, overcome due to illumination variation (suddenly or
Gradual change) and the influence of generation.
C) method of the Block- matching of variable block size from bottom to top more accurately determines possible object block, then adaptive
It answers Kalman filter for distinguishing the presence of significant movement and other distractive movements, improves correct detection target
Reliability detection and meanwhile reduce the presence of noise.
Detailed description of the invention
Fig. 1 is the flow chart of moving target detecting method of the invention.
Specific embodiment
For the ease of those of ordinary skill in the art understand and implement the present invention, below with reference to embodiment to the present invention make into
The detailed description of one step, it should be understood that implementation example described herein is merely to illustrate and explain the present invention, and is not used to limit
The fixed present invention.
The present invention provides a kind of moving target detecting method combined based on color characteristic with Kalman filter tracking.It is first
First, can be regarded for video processing problem as the problem of processing to each frame picture, and grayscale image can regard one as
Matrix xi∈Rw×h, wherein w is the width of image, and h is the height of image, and each element is some integer between 0~255 in matrix
Value, then can be converted into the processing of video to set of matrices X=x1, x2, x3..., xtProcessing, wherein t indicate view
The frame number of frequency.
The main thought of this method is: first calculating the color difference histogram (CDH) of pixel local neighborhood, then uses fuzzy c
Mean cluster (FCM) obtains fuzzy color difference histogram, and then carries out background modeling, and using before similitude matching detection
Scape further carries out target tracking using the Kalman filter based on Block- matching.
Algorithm flow chart of the invention is as shown in Figure 1, the specific steps are as follows:
Step 1: the color difference histogram of the local neighborhood of pixel is calculated using the color difference for corresponding to two color components
CDH。
1.1 calculating values of chromatism: all frames in video-frequency band are read in and are saved, are denoted as I respectively1, I2..., Im∈Nm×n,
Wherein M is the totalframes in video-frequency band, and m, n are the size of every frame picture;There to be the color image of strength grade I (p, q, ch)
It is quantified as W grade, such as I ∈ 0,1,2 ... W-1 };(p, q) is position coordinates, and ch is Color Channel.RGB image is turned
It is changed to CIE L*a*b* color model.
The color difference d in small part N × N neighborhood centered on position (p, q) is sought, is calculated using following formula:
Wherein I (r, s, ch) indicates the intensity of pixel in neighborhood.
1.2 color difference histograms: it will be blurred from (1) calculated color difference d using Gauss member function are as follows:
σ indicates standard deviation.
Color difference histogram is finally given by the following formula:
Wherein A × A is the regional area centered on the position (p, q), and the frequency of pixel is calculated with k.
Step 2: Fuzzy c-means Clustering
FCM is by K-bin local histogram H={ h1, h2, h3..., hkBe classified as with liC cluster centered on position, often
A cluster distributes bin based on fuzzy membership, this is completed by minimizing the iteration of cost function.
Fuzzy color difference histogram uses subordinated-degree matrix u (c × K dimension) and CDH vector H (K × 1) building c Victoria C DH (FCDH)
Fuzzy vector v, as follows:
hc×1=uc×KHK×1 (5)
Step 3: background subtraction method
3.1 background modelings: similarity measurement p is between modeling background frame F and present frame h using histogram measurement
The intersection of FCDH.
Wherein c is the quantity for providing histogram case.Background B is made comparisons by same threshold value and is obtained:
3.2 context updates: background FCDH uses [0,1] to update each pixel as Background learning rate, and t is time index,
It is as follows:
Step 4: the target tracking of the Kalman filter based on Block- matching
4.1 can be changed block size (BVBS) BM from bottom to top: just start that small size block is selected more to do more physical exercises to detect
Block,.Then block size doubles in each dimension.A series of pieces of { Bi, i=1,2,3,4 } } in smaller size, they
Share identical larger size father block Sb, it is defined as follows:
Sb=B1∪B2∪B3∪B4, where BI, l≤l≤4
1,2,3 and 4 father's block S is corresponded respectively tobUpper left, upper right, lower-left, bottom right sub-block.
B in step previousiDetecting the BM process with small moving mass can be repeated as desired for repeatedly, or until block
Until size reaches scheduled full-size.In this way, more false moving mass can be abandoned with biggish block size, keep simultaneously
The real motion details of smaller piece size.
4.2 times updated: known state xk-1, start iteration to NextState xkIt is predicted.
WhereinIt is prior estimate error covariance.Because the doing exercises for object block in X and Y coordinates is completely mutually
Independent, Kalman filtering can apply X, Y coordinates respectively.In each coordinate, the fast movement of a target can be with following
Formula indicates:
xk,ΔkRespectively indicate the displacement at k moment, speed, acceleration and the time difference.Because we do not know control
, the B in equationkukIt is 0.The specific equation of the motion tracking of real-time update is as follows:
The determination of process noise covariance is generally relatively difficult, because the estimation procedure directly observed is not always feasible
's.In this work, using acceleration as piecewise constant Wiener process, i.e. acceleration continues in each period for we
It is remained unchanged in time, but different for each period, and is each zero-mean gaussian sequence.
Q is that the energy spectral density density of continuous time white noise enables P to reduce calculation amount0=cQk, c is one normal
Number.
4.3: measurement updaue: state xkZkIt can be indicated with a linear equation:
zk=Hkxk+vk (13)
zkIt is the measured value of m × 1 at k moment, HkIt is and zkThe observing matrix of relevant m × n, vkIt is the additional survey of m × 1
Measure noise vector.The equation of specific measurement updaue is as follows:
Formula n × m matrix K is so that the smallest gain of posteriori error covariance.
In this work, measured value zkIt is most likely to be the target object block coordinate in current compensation frame.Therefore x,
Y, observing matrix H are in [100] in each direction definition.The position of object block is by three steps (TSS) BM algorithm measurement.Measurement is made an uproar
Sound vkIndicate the noise characteristic of sensor.Because only from the coordinate of the angle measurement target object of Block- matching, covariance matrix R exists
vkIn byScalar valueIt is that experience determines.
It should be understood that the part that this specification does not elaborate belongs to the prior art.
It should be understood that the above-mentioned description for preferred embodiment is more detailed, can not therefore be considered to this
The limitation of invention patent protection range, those skilled in the art under the inspiration of the present invention, are not departing from power of the present invention
Benefit requires to make replacement or deformation under protected ambit, fall within the scope of protection of the present invention, this hair
It is bright range is claimed to be determined by the appended claims.
Claims (5)
1. the moving target detecting method combined based on fuzzy means clustering with Kalman filter tracking, which is characterized in that institute
The method of stating includes the following steps:
Step 1, the color difference histogram CDH of pixel local neighborhood is calculated;
Step 2, fuzzy color difference histogram is obtained using Fuzzy c-means Clustering FCM;
Step 3, background modeling is carried out, and uses similitude matching detection prospect;
Step 4, target tracking is carried out using the Kalman filter based on Block- matching.
2. the moving object detection according to claim 1 combined based on fuzzy means clustering with Kalman filter tracking
Method, which is characterized in that step 1 is straight come the color difference for calculating the local neighborhood of pixel using the color difference for corresponding to two color components
Side figure CDH, specifically includes;
1.1 calculating values of chromatism: all frames in video-frequency band are read in and are saved, are denoted as I respectively1, I2..., Im∈Nm×n,
In, M is the totalframes in video-frequency band, and m, n are the size of every frame picture;There to be the color image of strength grade I (p, q, ch)
It is quantified as W grade, I ∈ 0,1,2 ... W-1 };(p, q) is position coordinates, and ch is Color Channel, and RGB image is converted to
CIE L*a*b* color model;
The color difference d in small part N × N neighborhood centered on position (p, q) is sought, is calculated using following formula:
Wherein, I (r, s, ch) indicates the intensity of pixel in neighborhood;
1.2 color difference histograms: it will be blurred from (1) calculated color difference d using Gauss member function are as follows:
σ indicates standard deviation;
Color difference histogram is finally given by the following formula:
Wherein, A × A is the regional area centered on the position (p, q), and the frequency of pixel is calculated with k.
3. the moving object detection according to claim 2 combined based on fuzzy means clustering with Kalman filter tracking
Method, which is characterized in that step 2 obtains fuzzy color difference histogram using Fuzzy c-means Clustering FCM, specifically includes:
FCM is by K-bin local histogram H={ h1, h2, h3..., hkBe classified as with liC cluster centered on position, Mei Geju
Class is based on fuzzy membership and distributes bin, this is completed by minimizing the iteration of cost function;
Fuzzy color difference histogram is fuzzy using subordinated-degree matrix u (c × K dimension) and CDH vector H (K × 1) building c Victoria C DH (FCDH)
Vector v is as follows:
hc×1=uc×KHK×1 (5)。
4. the moving object detection according to claim 3 combined based on fuzzy means clustering with Kalman filter tracking
Method, which is characterized in that step 3 carries out background modeling, and uses similitude matching detection prospect, specifically includes;
3.1 background modelings: similarity measurement p is FCDH between modeling background frame F and present frame h using histogram measurement
Intersection;
Wherein, c is the quantity for providing histogram case, and background B is made comparisons by same threshold value and obtained:
3.2 context updates: background FCDH uses [0,1] to update each pixel as Background learning rate, and t is time index, as follows
It is shown:
5. the moving object detection according to claim 4 combined based on fuzzy means clustering with Kalman filter tracking
Method, which is characterized in that step 4 carries out target tracking using the Kalman filter based on Block- matching, specifically includes:
4.1 can be changed block size (BVBS) BM from bottom to top: just start to select small size block to detect more moving mass, so
Block size doubles in each dimension afterwards, a series of pieces of { Bi, i=1,2,3,4 } } in smaller size, their shared phases
Same larger size father's block Sb, it is defined as follows:
Sb=B1∪B2∪B3∪B4, where BI, 1≤1≤4
1,2,3 and 4 father's block S is corresponded respectively tobUpper left, upper right, lower-left, bottom right sub-block;
B in step previousiDetecting the BM process with small moving mass can be repeated as desired for repeatedly, or until block size
Until reaching scheduled full-size, in this way, more false moving mass can be abandoned with biggish block size, while keeping smaller
The real motion details of block size;
4.2 times updated: known state xk-1, start iteration to NextState xkIt is predicted;
Wherein,It is prior estimate error covariance, because the doing exercises for object block in X and Y coordinates is completely mutually solely
Vertical, Kalman filtering can apply X, Y coordinates respectively, and in each coordinate, the fast movement of a target can use following public affairs
Formula indicates:
xk,Δ k respectively indicates the displacement at k moment, speed, acceleration and the time difference, because not knowing control item, equation
In BkukIt is 0, the specific equation of the motion tracking of real-time update is as follows:
The determination of process noise covariance is generally relatively difficult because the estimation procedure directly observed be not always it is feasible,
It, can be using acceleration as piecewise constant Wiener process, i.e. duration of the acceleration in each period in this work
It inside remains unchanged, but different for each period, and is each zero-mean gaussian sequence,
Q is that the energy spectral density density of continuous time white noise enables P to reduce calculation amount0=cQk, c is a constant;
4.3: measurement updaue: state xkZkIt can be indicated with a linear equation:
zk=Hkxk+vk (13)
zkIt is the measured value of m × 1 at k moment, HkIt is and zkThe observing matrix of relevant m × n, vkIt is the additional measurement noise of m × 1
The equation of vector, specific measurement updaue is as follows:
Formula n × m matrix K is so that the smallest gain of posteriori error covariance.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810549503.2A CN109102520A (en) | 2018-05-31 | 2018-05-31 | The moving target detecting method combined based on fuzzy means clustering with Kalman filter tracking |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810549503.2A CN109102520A (en) | 2018-05-31 | 2018-05-31 | The moving target detecting method combined based on fuzzy means clustering with Kalman filter tracking |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109102520A true CN109102520A (en) | 2018-12-28 |
Family
ID=64796582
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810549503.2A Pending CN109102520A (en) | 2018-05-31 | 2018-05-31 | The moving target detecting method combined based on fuzzy means clustering with Kalman filter tracking |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109102520A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110009660A (en) * | 2019-03-06 | 2019-07-12 | 浙江大学 | Object space method for tracing based on correlation algorithm filter |
CN110543881A (en) * | 2019-09-16 | 2019-12-06 | 湖北公众信息产业有限责任公司 | Video data management method based on cloud platform |
CN110910421A (en) * | 2019-11-11 | 2020-03-24 | 西北工业大学 | Weak and small moving object detection method based on block characterization and variable neighborhood clustering |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2009048112A1 (en) * | 2007-10-09 | 2009-04-16 | National University Corporation Hokkaido University | Dynamic image decoder, dynamic image decoding method, and dynamic image decoding program |
CN106951829A (en) * | 2017-02-23 | 2017-07-14 | 南京邮电大学 | A kind of notable method for checking object of video based on minimum spanning tree |
-
2018
- 2018-05-31 CN CN201810549503.2A patent/CN109102520A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2009048112A1 (en) * | 2007-10-09 | 2009-04-16 | National University Corporation Hokkaido University | Dynamic image decoder, dynamic image decoding method, and dynamic image decoding program |
CN106951829A (en) * | 2017-02-23 | 2017-07-14 | 南京邮电大学 | A kind of notable method for checking object of video based on minimum spanning tree |
Non-Patent Citations (2)
Title |
---|
XUESONG LE 等: "《Hybrid Salient Motion Detection Using Temporal Differencing and Kalman Filter Tracking with Non-Stationary Camera》", 《ICIP2017》 * |
康登飞 等: "《基于颜色差分直方图的运动目标提取》", 《微型机与应用》 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110009660A (en) * | 2019-03-06 | 2019-07-12 | 浙江大学 | Object space method for tracing based on correlation algorithm filter |
CN110543881A (en) * | 2019-09-16 | 2019-12-06 | 湖北公众信息产业有限责任公司 | Video data management method based on cloud platform |
CN110910421A (en) * | 2019-11-11 | 2020-03-24 | 西北工业大学 | Weak and small moving object detection method based on block characterization and variable neighborhood clustering |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Zhang et al. | A vehicle detection algorithm based on three-frame differencing and background subtraction | |
CN108876820B (en) | Moving target tracking method under shielding condition based on mean shift | |
CN111260738A (en) | Multi-scale target tracking method based on relevant filtering and self-adaptive feature fusion | |
CN107316321B (en) | Multi-feature fusion target tracking method and weight self-adaption method based on information entropy | |
CN107742306B (en) | Moving target tracking algorithm in intelligent vision | |
García et al. | Adaptive multi-cue 3D tracking of arbitrary objects | |
CN101916446A (en) | Gray level target tracking algorithm based on marginal information and mean shift | |
CN107067416A (en) | A kind of detection method of moving target | |
CN104167006B (en) | Gesture tracking method of any hand shape | |
CN108320300A (en) | A kind of space-time context visual tracking method of fusion particle filter | |
CN109102520A (en) | The moving target detecting method combined based on fuzzy means clustering with Kalman filter tracking | |
CN109166137A (en) | For shake Moving Object in Video Sequences detection algorithm | |
CN104036526A (en) | Gray target tracking method based on self-adaptive window | |
CN107169439A (en) | A kind of Pedestrians and vehicles detection and sorting technique | |
Kim et al. | Object Modeling with Color Arrangement for Region‐Based Tracking | |
CN111626107B (en) | Humanoid contour analysis and extraction method oriented to smart home scene | |
Jeyakar et al. | Robust object tracking using local kernels and background information | |
CN106372650B (en) | A kind of compression tracking based on motion prediction | |
CN106447685B (en) | A kind of infrared track method | |
Yu et al. | A crowd flow estimation method based on dynamic texture and GRNN | |
Qu et al. | Real-time traffic vehicle tracking based on improved MoG background extraction and motion segmentation | |
CN110738684A (en) | target tracking method based on correlation filtering fusion convolution residual learning | |
Xiao et al. | Research on moving object detection and tracking | |
CN111627047A (en) | Underwater fish dynamic visual sequence moving target detection method | |
Soeleman et al. | Tracking Moving Objects based on Background Subtraction using Kalman Filter |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181228 |
|
RJ01 | Rejection of invention patent application after publication |