CN110796676A - Target tracking method combining high-confidence updating strategy with SVM (support vector machine) re-detection technology - Google Patents
Target tracking method combining high-confidence updating strategy with SVM (support vector machine) re-detection technology Download PDFInfo
- Publication number
- CN110796676A CN110796676A CN201910959687.4A CN201910959687A CN110796676A CN 110796676 A CN110796676 A CN 110796676A CN 201910959687 A CN201910959687 A CN 201910959687A CN 110796676 A CN110796676 A CN 110796676A
- Authority
- CN
- China
- Prior art keywords
- target
- confidence
- response
- frame
- tracked
- 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
- 238000000034 method Methods 0.000 title claims abstract description 32
- 238000001514 detection method Methods 0.000 title claims abstract description 20
- 238000005516 engineering process Methods 0.000 title claims abstract description 16
- 238000012706 support-vector machine Methods 0.000 title abstract description 19
- 230000004044 response Effects 0.000 claims abstract description 63
- 238000004364 calculation method Methods 0.000 claims abstract description 23
- 238000001914 filtration Methods 0.000 claims abstract description 10
- 238000012545 processing Methods 0.000 claims description 11
- 238000010586 diagram Methods 0.000 claims description 8
- 239000011159 matrix material Substances 0.000 claims description 8
- 239000013598 vector Substances 0.000 claims description 8
- 230000010355 oscillation Effects 0.000 claims description 7
- 230000008859 change Effects 0.000 claims description 5
- 238000000605 extraction Methods 0.000 claims description 4
- 230000017105 transposition Effects 0.000 claims description 2
- 230000004807 localization Effects 0.000 abstract description 2
- 230000008569 process Effects 0.000 abstract description 2
- 238000011156 evaluation Methods 0.000 description 3
- 241000195940 Bryophyta Species 0.000 description 2
- 230000033001 locomotion Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- 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
- G06T7/251—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments involving models
-
- 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 invention discloses a target tracking method combining a high-confidence updating strategy and an SVM (support vector machine) re-detection technology, which is used for quickly positioning the possible position of a target based on a kernel-related filtering framework in the tracking process; the high confidence update module utilizes a filter response profile obtained from the target localization. And carrying out confidence judgment on the tracked target by using the highest response peak value of the response map and the two indexes of the fluctuation condition of the response map. And when the two judgment indexes are simultaneously higher than the set threshold value, updating the filter template, and in the target re-detection module, starting the detector when the confidence coefficient of the tracked target is judged to be too low. And detecting the possible positions of the target again and scoring, wherein the position with the highest score is the position of the tracked target. And re-detecting the condition that the target is blocked for a long time or is lost by using a re-detection module. The performance and the calculation speed of the target tracking algorithm are improved, and the target is stably tracked for a long time.
Description
Technical Field
The invention discloses a target tracking method combining a high-confidence updating strategy and an SVM (support vector machine) re-detection technology, and belongs to the technical field of computer vision and digital image processing.
Background
The target tracking technology is an important branch of computer vision, is widely applied to the fields of human-computer interaction, video monitoring, unmanned aerial vehicle investigation and the like, and is paid more and more attention to target detection and tracking along with the development of computer vision and artificial intelligence. However, since the tracked object may be in a complex environment, there are various factors such as illumination change, rigid deformation, fast movement, partial occlusion, and complicated background in the application, it is always a challenge to find a fast and stable tracking method suitable for object change.
In 2010, Blome et al propose an mosse (minimum Output Sum of Squared error filter) algorithm, which introduces correlation filtering into the target tracking field for the first time, and when calculating the correlation between a target and a sample set to be measured, applies fourier transform to directly transfer the operation into a frequency domain, thereby greatly reducing the amount of operation. Subsequently, Henriques et al proposed a CSK (expanding the Crcument Structure of Tracking-by-detection with Kernels) method using a circulant matrix Structure and kernel-space mapping based on MOSSE. The method only needs to extract a sample to be detected at a target position once, and then circularly shifts the sample to form a circular matrix. In 2014, Henriques et al introduced a kernel method based on CSK and adopted a multi-dimensional feature of hog (histogram of oriented gradients) to describe the appearance of a target (KCF), which greatly improves the robustness of relevant filtering target tracking.
For the kernel correlation filter tracking algorithm, the tracker (KCF is taken as an example hereinafter) adopts updating every frame during model updating, which is simple, but is difficult to adapt to complex situations such as rapid motion of a target, partial occlusion, background clutter and the like, and is easy to cause model drift to cause tracking failure. When the target is severely shielded or significantly deformed, the prediction result of the algorithm becomes unreliable, and the filter model is polluted. Therefore, the reliability of the tracking result needs to be verified, since it affects the subsequent model update.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides the target tracking method combining high-confidence updating and SVM redetection technology, which can improve the performance and the calculation speed of a target tracking algorithm and realize the long-time stable tracking of the target.
The invention is realized by the following technical scheme:
the high-confidence updating strategy is combined with a target tracking method of an SVM (support vector machine) re-detection technology, the confidence of a tracked target is judged through the high-confidence updating strategy, when the target is in a high-confidence state, a filter is updated, and when the target is in a low-confidence state, the target is re-detected; the method specifically comprises the following steps:
step 1: reading in a video frame sequence to be tracked, selecting an initial tracking target to be tracked on a first frame mark, initializing a tracker, and determining a target center candidate area;
step 2: extracting the characteristics of the search area, obtaining a correlation filtering response graph by using a kernel correlation filtering method, wherein the peak position in the graph is used as the position where a new frame of target response is possible to be located;
and step 3: carrying out confidence judgment on the target response, and updating the filter template when the target is in a high-confidence state; when the target is in a low confidence state, temporarily stopping updating the filter template;
and 4, step 4: when the target is in a low confidence state for a long time, judging that the target is blocked or the tracked target is lost; starting an SVM re-detection module to find the position of the tracked target again;
and 5: calculating a filter template according to the position of the target obtained by redetection, and carrying out interpolation updating on the tracked target template;
step 6: and (5) processing the next frame of image, and repeating the steps 2-5 until the video is finished.
Preferably, the step 1 is specifically realized by the following steps:
step 1.1: determining the central position of a tracked target in a given initial target frame;
step 1.2: and extracting HOG characteristics of the target center candidate region of the first frame, multiplying the obtained characteristics by a Hanning window to obtain a target template, and calculating a Gaussian matrix map according to the obtained template.
In the step 1, the size of the target center candidate region is 1-2 times, preferably 1.5 times, of the initial target bounding rectangle.
Preferably, the step 2 is specifically realized by the following steps:
step 2.1: and (3) extracting the features of the candidate region patch, and then performing Fourier transform to obtain the features of the tracked target, wherein the specific calculation expression is as follows:
in the formula, xf represents the feature extracted from the candidate region, F represents the Fourier transform, HOG represents the HOG feature extraction, and gauss represents the Gaussian response;
step 2.2: processing the next frame of region to be detected, extracting HOG characteristics of the tracked target, and performing Hanning window processing on the HOG characteristics, wherein the specific calculation expression is as follows:
in the formula, zf represents the feature of the detection region of the next frame, F represents fourier transform, HOG represents HOG feature extraction, and cos _ w represents the addition of a hanning window to the feature.
Step 2.3: and calculating the peak position of the response graph according to the extracted features, wherein the formula adopted by the calculation is as follows:
Fmax=max f(z)=max(response),
wherein F (z) represents a characteristic response, FmaxRepresents the maximum value of the characteristic response, and response represents the characteristic response.
Preferably, the step 3 is realized by the following steps:
when the target tracking is accurate, the peak value of the response graph is obvious and is close to the ideal two-dimensional Gaussian distribution; when the target is lost, the response graph vibrates; therefore, the target tracking reliability is judged according to the peak value and the oscillation condition of the target tracking response diagram, and the concrete calculation formula of the response diagram oscillation judgment is as follows:
in the formula: fmaxIs the maximum value of the target response, FminIs the minimum of the target response, Fw,hThe value of the target in the row and the column of the response matrix w;
only when the target tracking confidence index FmaxAnd EAPCEWhen the historical mean value exceeds the current frame by a certain proportion, the target response of the current frame is determined to have high confidence level, and the specific calculation formula is as follows:
where α is the template update confidence parameter, FmaxIs the peak response maximum, EAPCEFor the target confidence value, i and j refer to the current video frame, n refers to the total number of video frames, and a certain proportion in the algorithm is that α -0.45 and β -0.6 are used as conditions.
Preferably, the step 4 is realized by the following steps:
judging the confidence coefficient of the tracked target, namely judging the confidence coefficient E of the targetAPCE5 continuous frames are lower than the set credibility threshold TrAnd then, starting the detector, wherein the specific calculation formula is as follows:
in the formula (f)iRepresenting sample feature vectors, viIs the label of the characteristic vector, w is the solved hyperplane normal vector, w' is the transposition of the hyperplane normal vector, lambda is the regularization coefficient, and N is the total number of samples;
and scoring the predicted positions according to the formula, wherein the position with the highest score is the position predicted by the detector.
Preferably, the step 5 is realized by the following steps: calculating new ones based on the positions obtained by the detectorTemplate x 'obtained from frame target response position'tAnd parameter α't. And carrying out interpolation updating on the parameters of the current frame to enable the filter to adapt to the change of the target and improve the robustness of the tracker, wherein the specific calculation formula is as follows:
in the formula, αtFor the target filter parameters of the t frames after interpolation update, η is model update interpolation η is 0.01, xtFor interpolated updated t-frame model features, α'tTarget Filter parameters, x 'for the t-th frame'tThe model features are found for the t-th frame.
Compared with the prior art, the invention has the following beneficial effects:
the invention provides a target tracking method combining a high-confidence updating strategy and an SVM (support vector machine) re-detection technology on the basis of image processing, and aims to solve the problems that a tracked target is lost and a model is subjected to tracking drift when the tracked target is shielded by an obstacle.
The method respectively designs a target tracking high-confidence-degree updating module and an SVM re-detection technology module. In the tracking process, the position where the target is possibly located is quickly positioned based on the frame of kernel-dependent filtering; the high confidence update module utilizes a filter response profile obtained from the target localization. And carrying out confidence judgment on the tracked target by using the highest response peak value of the response map and the two indexes of the fluctuation condition of the response map. When the two judgment indexes are simultaneously higher than the set threshold value, the filter template is updated, so that the problem of model drift caused by the introduction of background noise and the updating of the filter template under the condition of low confidence coefficient is solved. And in the target re-detection module, when the confidence coefficient of the tracking target is judged to be too low, starting the detector. And detecting the possible positions of the target again and scoring, wherein the position with the highest score is the position of the tracked target. A re-detection module is utilized. And re-detecting the condition that the target is blocked for a long time or the target is lost. The invention realizes high robustness and long-time real-time tracking, and can track the target in time when the target is shielded or lost, thereby realizing real-time online tracking.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
FIG. 2 is a confidence level distribution and a peak fluctuation distribution graph of the algorithm of the present invention when the tracked target is occluded.
Fig. 3 is a one-pass success map of the algorithm of the present invention compared to other algorithms.
FIG. 4 is a one-pass accuracy graph of the algorithm of the present invention compared to other algorithms.
Detailed Description
The present invention will be described in further detail with reference to specific examples, but the scope of the present invention is not limited to these examples, and all changes or equivalent substitutions that do not depart from the spirit of the present invention are intended to be included within the scope of the present invention.
The target tracking method combining the high-confidence updating and SVM redetection technology specifically comprises the following steps:
step 1.1: at a given initial target frame, the center position of the tracked target is determined. The target center candidate region size is 1.5 times the original target bounding rectangle.
Step 1.2: and extracting HOG characteristics of the target center candidate region of the first frame. And multiplying the obtained characteristics by a Hanning window to obtain a target template, and calculating a Gaussian matrix diagram according to the obtained template.
Step 2, extracting the characteristics of the target search area of the new frame, and obtaining a relevant filtering response image by using a kernel relevant filtering method, wherein the peak position of the response image is possible to be the possible position of the target of the new frame;
step 2.1: extracting the characteristics of the search area (patch), then carrying out Fourier transform to obtain the characteristics of the tracked target, wherein the specific calculation expression is,
step 2.2: processing the region to be detected of the next frame to extract characteristics and carrying out Hanning window processing on the region to be detected of the next frame, wherein the specific calculation expression is,
step 2.3: calculating a response map peak location from the extracted features
Fmax=max f(z)=max(response)。
Step 3, carrying out confidence judgment on the target response obtained in the step 2, updating the filter template when the target is in a high-confidence state, and storing the extracted features and the label of the frame as reliability results into an SVM judgment set; when the target is in a low confidence state, temporarily stopping updating the filter template;
when the target tracking is accurate, the peak value of the response graph is obvious and is close to the ideal two-dimensional Gaussian distribution; when the target is lost, the response graph vibrates. Therefore, the target tracking reliability is judged according to the peak value and the oscillation condition of the target tracking response diagram. The concrete calculation formula of response diagram oscillation judgment is as follows:
in the formula: fmaxAnd FminThe maximum and minimum values of the target response, respectively. Fw,hThe values targeted at row and column of the response matrix w.
Only when the target tracking confidence index FmaxAnd EAPCEWhen the historical mean value exceeds the current frame by a certain proportion, the current frame can be determined to have high confidence level, and the specific calculation formula is as follows:
wherein the α template updates the confidence parameter.
And 4, when the target is in a low confidence state for a long time, judging that the target is blocked or the tracked target is lost. And starting an SVM re-detection module, and searching the closest numerical value of the features extracted by the frame and the SVM judgment set. The highest scoring position is the position where the tracked target is found again;
judging the confidence coefficient of the tracked target, namely judging the confidence coefficient E of the targetAPCE5 continuous frames are lower than the set credibility threshold TrThe detector is activated. The specific calculation formula is as follows:
and (4) scoring the predicted position through a formula, wherein the highest scoring position is the position predicted by the detector.
Step 5, calculating a template x obtained by a new frame according to the position obtained by the detectort-1And parameter αt-1. And the parameters of the current frame are interpolated and updated, so that the filter can adapt to the change of the target and the robustness of the tracker is improved.
The specific calculation formula is as follows:
and 6, processing the next frame of image, and repeating the step 2 until the video is finished.
As shown in FIG. 2, when the target is normally tracked, and the target is partially or completely occluded, the target response peak and the fluctuation situation distribution map. When the target tracking is accurate, the peak value of the response diagram is obvious and close to the ideal two-dimensional Gaussian distribution, as shown in FIG. 2 (a); when the target is partially shielded, the peak value F of the response graphmaxSharply reduced peak fluctuation determination index EAPCEAnd decreases as shown in fig. 2 (b). When the target is completely occluded, the peak value F of the response mapmaxAnd peak fluctuation condition determination index EAPCEBecomes very small as shown in fig. 2 (c). Therefore, the target tracking credibility is judged according to the peak value and the oscillation condition of the target tracking response image, and only when the peak value F of the response imagemaxAnd peak fluctuation condition determination index EAPCEIn a certain ratioIf the example is larger than the historical mean value, the tracked target can be determined to be in a high-confidence-degree state at the moment, and the template is updated.
As shown in FIG. 3, the chart heading "Success plots of OPE" is the one-pass evaluation Success rate, the chart abscissa "Overlap threshold" is the different threshold values, and the chart ordinate "Success rates" is the Success rate value. The chart label "Our" is the method of the present invention, "TLD", "Struck", "DSST", "KCF", and "MIL" are several other tracking algorithms that are currently popular. Under the condition that the algorithm provided by the invention passes the standard of the success rate evaluation once, the success rate of the 'Our' is 72.7%, which is far higher than the second algorithm 'DSST' 66.8%, and is nearly 6%, and the success rates of other algorithms are respectively as follows: 61.9% of KCF, 54.6% of Struck, 52.4% of TLD and 35.6% of MIL. Compared with other tracking algorithms, the method has good tracking performance.
As shown in FIG. 4, the chart heading "Precision plots of OPE" is the one-pass evaluation Precision, the chart abscissa "Location error threshold" is the local error threshold, and the chart ordinate "Precision" is the Precision value. Under the condition that the algorithm provided by the invention passes the standard of evaluating the accuracy rate once, the accuracy rate of the 'Our' is 84.7%, which is far higher than 74.7% of the second algorithm 'KCF', and is nearly 10 percentage points, and the success rates of other algorithms are respectively as follows: 74.4% for "DSST", "65.6% for" Struck "," 61.6% for "TLD", and 47.5% for "MIL". Compared with other tracking algorithms, the method has the advantages that the accuracy is greatly improved, and the tracked target can be accurately positioned under the condition that the target is shielded.
And aiming at the problems of partial or complete shielding, size conversion and the like existing in the target tracking video, the high-confidence strategy is used for judging the confidence level, and when the target is judged to be possibly shielded or lost, the target is re-detected. The method improves the performance and the calculation speed of the target tracking algorithm, can realize the long-time stable tracking of the target, effectively overcomes the interference of some environmental factors, and has wide practical engineering application value.
The present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (7)
1. The target tracking method is characterized in that the confidence of a tracked target is judged through a high-confidence updating strategy, when the target is in a high-confidence state, a filter is updated, and when the target is in a low-confidence state, the target is re-detected; the method specifically comprises the following steps:
step 1: reading in a video frame sequence to be tracked, selecting an initial target to be tracked on a first frame mark, initializing a tracker, and determining a target center candidate region;
step 2: extracting the characteristics of the candidate region, obtaining a correlation filtering response graph by using a kernel correlation filtering method, wherein the peak position in the graph is used as the position where a new frame of target response is possible to be located;
and step 3: carrying out confidence judgment on the target response, and updating the filter template when the target is in a high-confidence state; when the target is in a low confidence state, temporarily stopping updating the filter template;
and 4, step 4: when the target is in a low confidence state for a long time, judging that the target is blocked or the tracked target is lost; starting an SVM re-detection module to find the position of the tracked target again;
and 5: calculating a filter template according to the position of the target obtained by redetection, and carrying out interpolation updating on the tracked target template;
step 6: and (5) processing the next frame of image, and repeating the steps 2-5 until the video is finished.
2. The target tracking method based on the combination of the high-confidence updating strategy and the SVM re-detection technology according to claim 1, wherein the step 1 is specifically realized by the following steps:
step 1.1: determining the central position of a tracked target in a given initial target frame;
step 1.2: and extracting HOG characteristics of the target center candidate region of the first frame, multiplying the obtained characteristics by a Hanning window to obtain a target template, and calculating a Gaussian matrix map according to the obtained template.
3. The method for tracking the target by combining the high-confidence updating strategy and the SVM redetection technology as claimed in claim 1, wherein in the step 1, the size of the target center candidate region is 1-2 times of the initial target bounding rectangle.
4. The target tracking method based on the combination of the high-confidence updating strategy and the SVM re-detection technology as claimed in claim 1, wherein the step 2 is specifically implemented by:
step 2.1: and (3) extracting the features of the candidate region patch, and then performing Fourier transform to obtain the features of the tracked target, wherein the specific calculation expression is as follows:
in the formula, xf represents the feature extracted from the candidate region, F represents the Fourier transform, HOG represents the HOG feature extraction, and gauss represents the Gaussian response;
step 2.2: processing the next frame of region to be detected, extracting HOG characteristics of the tracked target, and performing Hanning window processing on the HOG characteristics, wherein the specific calculation expression is as follows:
in the formula, zf represents the feature of a detection area of the next frame, F represents Fourier transform, HOG represents HOG feature extraction, and cos _ w represents the addition of a Hanning window to the feature;
step 2.3: calculating the peak position of the response graph according to the extracted HOG characteristics, wherein the formula adopted by calculation is as follows:
Fmax=maxf(z)=max(response),
wherein F (z) represents a characteristic response, FmaxRepresents the maximum value of the characteristic response, and response represents the characteristic response.
5. The target tracking method based on the combination of the high-confidence updating strategy and the SVM redetection technology in claim 1 is characterized in that the step 3 is realized by the following steps:
judging the target tracking credibility according to the peak value and the oscillation condition of the target tracking response diagram, wherein the concrete calculation formula of response diagram oscillation judgment is as follows:
in the formula: fmaxIs the maximum value of the target response, FminIs the minimum of the target response, Fw,hThe value of the target in the row and the column of the response matrix w;
only when the target tracking confidence index FmaxAnd EAPCEWhen the historical mean value exceeds the current frame by a certain proportion, the target response of the current frame is determined to have high confidence level, and the specific calculation formula is as follows:
where α is the template update confidence parameter, FmaxIs the peak response maximum, EAPCEFor the target confidence value, i and j refer to the current video frame, n refers to the total number of video frames, and a certain proportion in the algorithm is that α -0.45 and β -0.6 are used as conditions.
6. The target tracking method based on the combination of the high-confidence updating strategy and the SVM redetection technology in claim 1 is characterized in that the step 4 is realized by the following steps:
judging the confidence coefficient of the tracked target, namely judging the confidence coefficient E of the targetAPCE5 continuous frames are lower than the set credibility threshold TrThe detector is started, and the specific calculation formula is as follows:
in the formula (f)iRepresenting sample feature vectors, viIs the label of the characteristic vector, w is the solved hyperplane normal vector, w' is the transposition of the hyperplane normal vector, lambda is the regularization coefficient, and N is the total number of samples;
and scoring the predicted positions according to the formula, wherein the position with the highest score is the position predicted by the detector.
7. The method for tracking the target by combining the high-confidence updating strategy with the SVM redetection technology as claimed in claim 1, wherein the step 5 is implemented by:
calculating a template x 'obtained from a new frame of target response position according to the position obtained by the detector'tAnd parameter α'tAnd performing interpolation updating on the parameters of the current frame by combining the characteristics and the parameters of the model of the previous frame, so that the filter adapts to the change of the target and the robustness of the tracker is improved, wherein the specific calculation formula is as follows:
in the formula, αtFor the target filter parameters of t frames after interpolation update, η is model update interpolation, η is 0.01, xtFor interpolated updated t-frame model features, α'tTarget Filter parameters, x 'for the t-th frame'tThe model features are found for the t-th frame.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910959687.4A CN110796676A (en) | 2019-10-10 | 2019-10-10 | Target tracking method combining high-confidence updating strategy with SVM (support vector machine) re-detection technology |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910959687.4A CN110796676A (en) | 2019-10-10 | 2019-10-10 | Target tracking method combining high-confidence updating strategy with SVM (support vector machine) re-detection technology |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110796676A true CN110796676A (en) | 2020-02-14 |
Family
ID=69438982
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910959687.4A Pending CN110796676A (en) | 2019-10-10 | 2019-10-10 | Target tracking method combining high-confidence updating strategy with SVM (support vector machine) re-detection technology |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110796676A (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111696132A (en) * | 2020-05-15 | 2020-09-22 | 深圳市优必选科技股份有限公司 | Target tracking method and device, computer readable storage medium and robot |
CN111862160A (en) * | 2020-07-23 | 2020-10-30 | 中国兵器装备集团自动化研究所 | Target tracking method, medium and system based on ARM platform |
CN112150511A (en) * | 2020-11-02 | 2020-12-29 | 电子科技大学 | Target tracking algorithm based on combination of image matching and improved kernel correlation filter |
CN112164093A (en) * | 2020-08-27 | 2021-01-01 | 同济大学 | Automatic person tracking method based on edge features and related filtering |
CN112184764A (en) * | 2020-09-10 | 2021-01-05 | 太原理工大学 | Target tracking method based on Fourier-Mellin transform |
CN112561958A (en) * | 2020-12-04 | 2021-03-26 | 武汉华中天经通视科技有限公司 | Correlation filtering image tracking loss judgment method |
CN112699718A (en) * | 2020-04-15 | 2021-04-23 | 南京工程学院 | Scale and illumination self-adaptive structured multi-target tracking method and application thereof |
CN112884037A (en) * | 2021-02-09 | 2021-06-01 | 中国科学院光电技术研究所 | Target tracking method based on template updating and anchor-frame-free mode |
CN113537241A (en) * | 2021-07-16 | 2021-10-22 | 重庆邮电大学 | Long-term correlation filtering target tracking method based on adaptive feature fusion |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108830879A (en) * | 2018-05-29 | 2018-11-16 | 上海大学 | A kind of unmanned boat sea correlation filtering method for tracking target suitable for blocking scene |
-
2019
- 2019-10-10 CN CN201910959687.4A patent/CN110796676A/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108830879A (en) * | 2018-05-29 | 2018-11-16 | 上海大学 | A kind of unmanned boat sea correlation filtering method for tracking target suitable for blocking scene |
Non-Patent Citations (2)
Title |
---|
林彬 等: "基于高置信度更新策略的高速相关滤波跟踪算法" * |
阮宏刚: "基于核相关滤波的目标跟踪方法研究" * |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112699718A (en) * | 2020-04-15 | 2021-04-23 | 南京工程学院 | Scale and illumination self-adaptive structured multi-target tracking method and application thereof |
CN112699718B (en) * | 2020-04-15 | 2024-05-28 | 南京工程学院 | Scale and illumination self-adaptive structured multi-target tracking method and application thereof |
CN111696132A (en) * | 2020-05-15 | 2020-09-22 | 深圳市优必选科技股份有限公司 | Target tracking method and device, computer readable storage medium and robot |
CN111696132B (en) * | 2020-05-15 | 2023-12-29 | 深圳市优必选科技股份有限公司 | Target tracking method, device, computer readable storage medium and robot |
CN111862160A (en) * | 2020-07-23 | 2020-10-30 | 中国兵器装备集团自动化研究所 | Target tracking method, medium and system based on ARM platform |
CN111862160B (en) * | 2020-07-23 | 2023-10-13 | 中国兵器装备集团自动化研究所有限公司 | Target tracking method, medium and system based on ARM platform |
CN112164093A (en) * | 2020-08-27 | 2021-01-01 | 同济大学 | Automatic person tracking method based on edge features and related filtering |
CN112184764A (en) * | 2020-09-10 | 2021-01-05 | 太原理工大学 | Target tracking method based on Fourier-Mellin transform |
CN112150511A (en) * | 2020-11-02 | 2020-12-29 | 电子科技大学 | Target tracking algorithm based on combination of image matching and improved kernel correlation filter |
CN112561958A (en) * | 2020-12-04 | 2021-03-26 | 武汉华中天经通视科技有限公司 | Correlation filtering image tracking loss judgment method |
CN112884037A (en) * | 2021-02-09 | 2021-06-01 | 中国科学院光电技术研究所 | Target tracking method based on template updating and anchor-frame-free mode |
CN112884037B (en) * | 2021-02-09 | 2022-10-21 | 中国科学院光电技术研究所 | Target tracking method based on template updating and anchor-frame-free mode |
CN113537241A (en) * | 2021-07-16 | 2021-10-22 | 重庆邮电大学 | Long-term correlation filtering target tracking method based on adaptive feature fusion |
CN113537241B (en) * | 2021-07-16 | 2022-11-08 | 重庆邮电大学 | Long-term correlation filtering target tracking method based on adaptive feature fusion |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110796676A (en) | Target tracking method combining high-confidence updating strategy with SVM (support vector machine) re-detection technology | |
CN111127518B (en) | Target tracking method and device based on unmanned aerial vehicle | |
Feng et al. | Dynamic saliency-aware regularization for correlation filter-based object tracking | |
Xuan et al. | Object tracking in satellite videos by improved correlation filters with motion estimations | |
CN111260689A (en) | Effective confidence enhancement correlation filtering visual tracking algorithm | |
CN113327272B (en) | Robustness long-time tracking method based on correlation filtering | |
CN111274917B (en) | Long-time target tracking method based on depth detection | |
CN110555870A (en) | DCF tracking confidence evaluation and classifier updating method based on neural network | |
CN110276782B (en) | Hyperspectral target tracking method combining spatial spectral features and related filtering | |
CN117576380A (en) | Target autonomous detection tracking method and system | |
CN110428447B (en) | Target tracking method and system based on strategy gradient | |
CN116665097A (en) | Self-adaptive target tracking method combining context awareness | |
CN108805909B (en) | Target tracking method based on particle filter redetection under related filter framework | |
CN116012421A (en) | Target tracking method and device | |
CN111241965A (en) | Target tracking method for occlusion detection based on multi-example learning | |
CN112184764A (en) | Target tracking method based on Fourier-Mellin transform | |
US20230029495A1 (en) | Object identification and tracking method and apparatus | |
CN112614158B (en) | Sampling frame self-adaptive multi-feature fusion online target tracking method | |
CN115272409A (en) | Single-target long-time tracking method based on deep neural network | |
CN110660081B (en) | Target tracking method based on self-adaptive feature selection | |
CN113920155A (en) | Moving target tracking algorithm based on kernel correlation filtering | |
CN112465865A (en) | Multi-target tracking method based on background modeling and IoU matching | |
CN114897932B (en) | Infrared target tracking realization method based on feature and gray level fusion | |
Ogawa et al. | Reducing false positives in object tracking with Siamese network | |
CN111160190A (en) | Vehicle-mounted pedestrian detection-oriented classification auxiliary kernel correlation filtering tracking method |
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: 20200214 |
|
RJ01 | Rejection of invention patent application after publication |