CN114332157B - Long-time tracking method for double-threshold control - Google Patents
Long-time tracking method for double-threshold control Download PDFInfo
- Publication number
- CN114332157B CN114332157B CN202111527248.XA CN202111527248A CN114332157B CN 114332157 B CN114332157 B CN 114332157B CN 202111527248 A CN202111527248 A CN 202111527248A CN 114332157 B CN114332157 B CN 114332157B
- Authority
- CN
- China
- Prior art keywords
- network
- matching
- frame
- search area
- tracking
- 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
- 238000000034 method Methods 0.000 title claims abstract description 50
- 238000012795 verification Methods 0.000 claims abstract description 46
- 238000012549 training Methods 0.000 claims abstract description 9
- 230000004044 response Effects 0.000 claims description 12
- 230000009466 transformation Effects 0.000 claims description 11
- 238000000605 extraction Methods 0.000 claims description 8
- 238000013527 convolutional neural network Methods 0.000 claims description 7
- 238000011423 initialization method Methods 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 4
- 238000011156 evaluation Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 abstract description 13
- 230000000694 effects Effects 0.000 abstract description 6
- 230000007547 defect Effects 0.000 abstract description 3
- 230000007774 longterm Effects 0.000 description 9
- 230000008859 change Effects 0.000 description 7
- 238000013528 artificial neural network Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 230000000007 visual effect Effects 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 230000008034 disappearance Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The invention provides a long-time tracking method with double threshold control, which achieves good effect in different tracking scenes. The invention integrates the verification network and the twin network, adopts a double-threshold control method, and judges various conditions in the long-time tracking process so as to ensure long-time robustness of the algorithm. The invention fuses the two types of networks, well utilizes the advantages of the two types of networks and makes up the defects of the two types of networks so as to adapt to long-time tracking scenes. The MDNet-based verification network well utilizes information in subsequent frames through online training, and the problem of missing tracking target information in a tracking algorithm based on a twin network is solved; the tracking algorithm based on the twin network replaces the network prediction process in the tracking algorithm based on MDNet by a template matching method, so that the calculated amount is small, and the problem of poor real-time performance of the tracking algorithm based on MDNet is solved.
Description
Technical Field
The invention relates to the technical field of computer vision, in particular to a long-time tracking method for double-threshold control.
Background
In the technical field of computer vision, the target tracking technology is used as one of research hotspots, and has higher research value in a plurality of fields such as video monitoring, intelligent transportation, man-machine interaction, medical diagnosis and the like. The main workflow of the object tracking technique is to first give a specific position of the object in the first frame of the video sequence and then locate the same object in each subsequent frame by some algorithm. The tracking algorithm can be divided into short-time tracking and long-time tracking according to the tracking time. For short-time tracking, the target is always in the field of view of the camera, and the main problem of algorithm research is how to quickly and accurately locate the position of the target in the subsequent frame.
In recent years, in the field of short-time tracking, various algorithms are rapidly developed, and remarkable results are obtained in data sets such as OTB, UAV123 and the like. However, in the field of long-term tracking, due to uncertainty of target and camera motion, various short-time tracking algorithms are difficult to cope with various situations such as camera view angle change, long-term shielding of the target, and returning of the target to the field of view after the target removes the field of view. Meanwhile, compared with short-time tracking, long-time tracking is closer to practical application, and has higher research value.
In the short-time tracking field, two methods are mainly included: one class is a class-based validation method, such as a CNN-based class tracking method, and the like. The target appearance is learned online, and the target position is distinguished in the background, so that the tracking effect is achieved; the other type is a regression method based on matching, such as a twin network method, wherein the target data features of a first frame are extracted through an offline trained neural network, template matching is performed in a subsequent frame, and the optimal candidate region is selected as a target position, so that the tracking effect is achieved. However, for the online updated classified tracker, when the situation that the target in long-term tracking is blocked or the target removes the visual field is faced, the online updated classified tracker is easily interfered by noise, and the information of the background is updated in an error manner, so that the tracking algorithm is invalid. For the matching tracker for offline training, as only the target appearance information of the first frame is extracted, the tracking algorithm is easy to fail when the subsequent frame faces change the camera visual angle or the target appearance in long-term tracking.
At present, a neural network-based detector is mostly introduced under the conditions of target disappearance and the like of a long-time tracking framework, and the field of view is re-detected so as to relocate the target position. The scheme has the defects that firstly, the detector needs a large amount of data to perform offline training, and the generalization capability is poor, so that the workload of early preparation is increased; secondly, in order to keep good detection results, the network depth of the general detector is designed to be deeper, so that the calculated amount in the tracking process is increased, and the real-time performance is affected; thirdly, in the tracking process, because the network depth is deeper, on-line training is inconvenient to carry out according to the target characteristics of the current frame, and the change of the appearance of the tracked target cannot be adapted on line.
Therefore, a long-term tracking method is needed at present, which can integrate two algorithms of short-term tracking to achieve the effect of stable work in long-term tracking.
Disclosure of Invention
In view of the above, the invention provides a long-term tracking method controlled by double thresholds, which can integrate two algorithms of short-term tracking to achieve the effect of stable work in long-term tracking.
In order to achieve the above purpose, the technical scheme of the invention is as follows:
a long-time tracking method for double-threshold control comprises the following specific steps:
S1, constructing a double-network structure consisting of a twin network and an authentication network, wherein the twin network comprises a feature extraction network and an RPN network; and (5) performing initial frame calibration to obtain a target and a target position frame.
S2, initializing a twin network by utilizing a target position frame corresponding to the initial frame to obtain a search area and a matching template of the initial frame; and initializing the verification network and updating the verification network parameters.
S3 is performed with the initial frame as the current frame of the process.
S3, extracting multi-scale features of the current frame in the updated search area of the previous frame by utilizing a feature extraction network to obtain matching template features and search area features; the RPN performs matching operation according to the matching template characteristics and the search area characteristics to obtain a matching tracking result; and sending the matching tracking result into the verification network.
And S4, performing tracking state evaluation on the matching tracking result of the current frame by the verification network to obtain a tracking result score.
S5, setting template updating conditions and search area transformation conditions, and judging the tracking result score:
When the continuous N frames of the tracking result score are larger than the template updating threshold value, the current search area and the current matching template are not updated, a next frame is directly taken as the current frame to be processed, and S3 is returned.
When the continuous N frames of the tracking result scores are smaller than the template updating threshold and larger than the searching area updating threshold, the target positions corresponding to the matching tracking results are utilized to reinitialize the twin network and the verification network, update parameters of the verification network and update the current matching template; then, the next frame is selected as the current frame to be processed, and the process returns to S3.
When the continuous N frames of the tracking result scores are smaller than the search area transformation threshold, updating the search area to be global, carrying out global matching tracking in the global by utilizing a twin network, simultaneously, grading the global matching tracking result by a verification network, and changing the global search into the original search area size when the global matching score is higher than the search area transformation threshold; then, the next frame is selected as the current frame to be processed, and the process returns to S3.
And finishing processing all frames in the video to be processed, and ending the flow.
Further, initializing the twin network by utilizing the target position corresponding to the initial frame to obtain the search area and the matching template of the initial frame, initializing the verification network, and updating the verification network parameters, wherein the specific method comprises the following steps:
The initialization method of the twin network comprises the following steps: extracting characteristics of a target to be used as a matching template; the four-fold range around the target position frame is used as the search area of the initial frame.
The initialization method of the verification network comprises the following steps: randomly generating a position frame with the intersection ratio of more than 0.7 with the target position frame by using a Gaussian distribution method as a positive sample, and generating a position frame with the intersection ratio of less than 0.3 with the target position frame by using a uniform random method as a negative sample; and substituting the positive sample and the negative sample into the verification network for training, and initializing parameters of the verification network.
Further, the number of positive samples is 500, and the number of negative samples is 5000.
Further, the feature extraction network is utilized to extract multi-scale features in the search area updated in the previous frame, so as to obtain the matched template features and the search area features, and the specific method is as follows:
The matched template is subjected to convolutional neural network to obtain 6 multiplied by 256 matched template characteristics; the search area is subjected to convolutional neural network to obtain 22×22×256 search area characteristics.
Further, the RPN performs matching operation according to the matching template characteristics and the search area characteristics to obtain a matching tracking result, and the specific method comprises the following steps:
Matching the template features and the search area features, inputting the matched template features and the search area features into the classification branches of the RPN network, and obtaining a classification response graph.
Matching the template features and the search area features, inputting the matched template features and the search area features into regression branches of the RPN network, and obtaining a regression response graph.
And adjusting the position of the target according to the classification response graph and the regression response graph to obtain a matching tracking result.
The beneficial effects are that:
1. The invention provides a target tracking method based on a verification network and a twin network, which achieves good effects in different tracking scenes. The invention integrates the verification network and the twin network, and designs a complete set of framework to be applied to long-time target tracking. The invention adopts SiamRPN technology based on twin network to extract target characteristics and match templates; updating a template by adopting a verification network, and maximally separating background information and target information to obtain an accurate matching score; and a double-threshold control method is adopted to judge various conditions in the long-time tracking process so as to ensure the long-time robustness of the algorithm. The invention fuses the two types of networks, well utilizes the advantages of the two types of networks and makes up the defects of the two types of networks so as to adapt to long-time tracking scenes. The MDNet-based verification network well utilizes information in subsequent frames through online training, and the problem of missing tracking target information in a tracking algorithm based on a twin network is solved; the tracking algorithm based on the twin network replaces the network prediction process in the tracking algorithm based on MDNet by a template matching method, so that the calculated amount is small, and the problem of poor real-time performance of the tracking algorithm based on MDNet is solved.
2. According to the invention, by introducing the MDNet-based verification network, the problem that the common neural network cannot be updated on line due to large calculated amount is solved. Experiments prove that when the target disappears, the tracking failure caused by the return to the visual field is caused, and the main reason is not that the tracker cannot adapt to the change of the appearance of the target well. Instead, the uncertainty of the target reproduction position causes the target reproduction position to exceed the search range of the tracker, thereby causing the failure of the tracking algorithm. The present invention enlarges the search area in the face of the disappearance of the target by the double threshold control method, thereby replacing the reinspection operation in the general long-time tracking frame. The method avoids the introduction of the neural network, reduces the calculation amount and improves the tracking efficiency.
Drawings
FIG. 1 is a block diagram of an implementation of the present invention.
Fig. 2 is a block diagram of a twin network portion of the present invention.
Fig. 3 is a block diagram of the regression network portion of the present invention.
Detailed Description
The invention will now be described in detail by way of example with reference to the accompanying drawings.
As shown in fig. 1, the invention provides a long-term tracking method controlled by double threshold values, which comprises the following specific steps:
S1, constructing a double-network structure consisting of a twin network and a verification network, performing initial frame calibration on a video to be processed, and giving a target in an initial frame and a corresponding target position frame.
S2, initializing a twin network by utilizing a target position frame corresponding to the initial frame to obtain a search area and a matching template of the initial frame; and initializing the verification network and updating the verification network parameters.
In the embodiment of the invention, the main operation of twin network initialization is to extract the characteristics of the target, and a 6 multiplied by 256 image is obtained and is used as a matching template; according to the central position of the target, the four-time size range near the target position frame is used as the search area of the next frame, and the proper setting of the search area not only avoids the redundant calculation amount caused by global search, but also prevents the failure of the tracking algorithm caused by the fact that the target exceeds the search area.
The main operation of network initialization is that according to the position of the target in the initial frame, a Gaussian distribution method is used for randomly generating a position frame with the intersection ratio of the position frame and the target being more than 0.7 as a positive sample, and a uniform random method is used for generating a position frame with the intersection ratio of the position frame and the target being less than 0.3 as a negative sample; the positive sample and the negative sample are substituted into the verification network for training, and parameters of the network are initially verified, so that the network can adapt to the initial appearance of the target, and the target and the background in the distinction are maximized. In the embodiment of the invention, the number of the selected positive samples is 500, and the number of the selected negative samples is 5000. In the follow-up tracking, in the search area obtained in the previous frame, matching operation is carried out according to the initialized matching template, so as to obtain a matching tracking result of the frame, and the search area of the next frame is calculated.
And S3, extracting multi-scale features of the current frame in the updated search area of the previous frame by utilizing a feature extraction network to obtain matching template features and search area features. The matched template is subjected to convolutional neural network to obtain 6 multiplied by 256 matched template characteristics; the search area is subjected to convolutional neural network to obtain 22×22×256 search area characteristics.
As shown in fig. 2, the RPN network performs a matching operation according to the matching template features and the search region features to obtain a matching tracking result; and sending the matching tracking result into the verification network.
The matched template features are input into classification branches of the RPN network, and convolved by convolution check, so as to obtain template feature frames; the searching region features are input into classifying branches of the RPN network, and convolving is carried out on the searching region features through convolution check to obtain searching region frame features; and carrying out convolution processing on the frame features of the search area through a convolution kernel according to the template feature frame to obtain a 17 multiplied by 2k classification response diagram. The size of the template feature frame is 4×4× (2 k×256), and the size of the search area frame feature is 20×20×256. The meaning of k in the template characteristic frame is that k different changes exist in k different anchors corresponding to k different anchors, and k groups of foreground and background classified template characteristics are obtained.
Matching template features and search area features are input into regression branches of the RPN network, and the specific mode is the same as a processing method of classification branches, so that a 17 multiplied by 4k regression response graph is obtained, wherein 4 represents offset of four directions, namely fine adjustment of a tracking frame obtained through regression.
And adjusting the position of the target according to the classification response graph and the regression response graph to obtain a matching tracking result.
S4, as shown in FIG. 3, the verification network carries out tracking state evaluation on the matching tracking result of the current frame to obtain a tracking result score.
S5, setting template updating conditions and search area transformation conditions, and judging the tracking result score. In order to better judge the state of the current frame of the tracking target in the long-time tracking process and change the tracking operation according to the state, a double-threshold control method is introduced.
The first threshold is a template updating threshold, and in the long-time tracking process, the appearance of the target can change in angle due to factors such as an observation angle, target movement and the like. In experiments, when the appearance of a tracking target gradually happens, the twin network can adaptively track in a short time, but if the first frame template is used for matching tracking for a long time, the tracking frame can drift. A template update threshold is introduced to determine the point in time at which a template update is required. When the tracking score is less than the template update threshold, the target appearance may have changed at this point, and the tracking twin network template is updated while the verification network parameters are updated.
The second threshold is a search area transformation threshold, and in the long-time tracking process, the conditions of target shielding, target removal, visual field and the like are likely to occur, and at the moment, if the template is updated by mistake, the template is polluted to cause subsequent tracking failure. Experiments show that the appearance change of the target is not much different from that before the target is shielded under most conditions, and the matching can be successfully performed by virtue of generalization of the twin network. The main reason for algorithm failure is that the target reproduction position uncertainty often exceeds the target search area, so that the tracking algorithm is disabled, and a search area transformation threshold is introduced.
When the score of N continuous frames is larger than the template updating threshold value, the target appearance of the frame is not changed more than that of the initial frame, so that the matching tracking operation of the next frame is continued.
When the score of N continuous frames is smaller than the template updating threshold value and larger than the searching region transformation threshold value, the frame matching tracking result is accurate, but the appearance of the target is greatly different from the appearance of the initial frame at the moment, and tracking errors can be caused by matching tracking. Therefore, the twin network and the verification network are re-initialized at the moment, the specific operation is the same as that of the first frame initialization, the matching tracking result of the previous frame is used as the initialization input of the present frame, the matching template and the verification network parameters are updated in sequence, and the tracking is performed again after the updating is completed.
In the embodiment of the invention, unlike the tracking algorithm based on MDNet networks in general, the verification network is only used for performing verification operation of tracking results, and is not used for predicting the position information of the target. In the tracking process, a tracking result with the score being larger than the template updating threshold value is stored as a positive sample, and when the score is smaller than the template updating threshold value, network updating operation is carried out. The specific operation is that positive samples and negative samples are generated in the same way as the initialization operation, wherein the positive samples are selected from 100 frames which are successfully tracked recently, and the negative samples are generated from 20 frames which are successfully tracked recently. And the positive and negative samples are sent into a verification network for iterative training in a ratio of 1:3, the iteration times are 10 times, and the batch size is 128. After the iteration is completed, the updating operation of the verification network is completed, and the tracking result verification operation of the subsequent frames is continued.
When the score of N consecutive frames is smaller than the search area conversion threshold, it is indicated that a situation in which the target is blocked by an obstacle or the target removes the field of view may occur at this time. At this time, the search area is enlarged, and the previous search area is enlarged to the global search. And carrying out matching tracking in the whole graph range, grading by the verification network according to the matching result, changing the global search into the original search area size when the score is higher than the search area transformation threshold, and continuing the follow-up tracking operation. And finishing processing all frames in the video to be processed, and ending the flow.
In summary, the above embodiments are only preferred embodiments of the present invention, and are not intended to limit the scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (5)
1. The long-time tracking method for the double-threshold control is characterized by comprising the following specific steps of:
S1, constructing a double-network structure consisting of a twin network and an authentication network, wherein the twin network comprises a feature extraction network and an RPN network; performing initial frame calibration to obtain a target and a target position frame;
S2, initializing a twin network by utilizing a target position frame corresponding to the initial frame to obtain a search area and a matching template of the initial frame; initializing a verification network and updating verification network parameters;
s3, taking the initial frame as a current frame to be processed, and executing the step S3;
S3, extracting multi-scale features of the current frame in the updated search area of the previous frame by utilizing a feature extraction network to obtain matching template features and search area features; the RPN performs matching operation according to the matching template characteristics and the search area characteristics to obtain a matching tracking result; sending the matching tracking result into the verification network;
S4, the verification network carries out tracking state evaluation on the matching tracking result of the current frame to obtain a tracking result score;
s5, setting template updating conditions and search area transformation conditions, and judging the tracking result score:
When the continuous N frames of the tracking result score are larger than the template updating threshold value, the current searching area and the current matching template are not updated, a next frame is directly taken as a current frame to be processed, and S3 is returned;
When the continuous N frames of the tracking result scores are smaller than the template updating threshold and larger than the searching area updating threshold, the target positions corresponding to the matching tracking results are utilized to reinitialize the twin network and the verification network, update parameters of the verification network and update the current matching template; then, selecting the next frame as the current frame to be processed, and returning to the step S3;
When the continuous N frames of the tracking result scores are smaller than the search area transformation threshold, updating the search area to be global, carrying out global matching tracking in the global by utilizing a twin network, simultaneously, grading the global matching tracking result by a verification network, and changing the global search into the original search area size when the global matching score is higher than the search area transformation threshold; then, selecting the next frame as the current frame to be processed, and returning to the step S3;
and finishing processing all frames in the video to be processed, and ending the flow.
2. The method of claim 1, wherein the initializing the twin network with the target position corresponding to the initial frame to obtain the search area and the matching template of the initial frame, initializing the verification network, and updating the verification network parameters comprises the following specific steps:
the initialization method of the twin network comprises the following steps: extracting the characteristics of the target as a matching template; taking the four-time range around the target position frame as a searching area of an initial frame;
The initialization method of the verification network comprises the following steps: randomly generating a position frame with the intersection ratio of more than 0.7 with the target position frame by using a Gaussian distribution method as a positive sample, and generating a position frame with the intersection ratio of less than 0.3 with the target position frame by using a uniform random method as a negative sample; and substituting the positive sample and the negative sample into the verification network for training, and initializing parameters of the verification network.
3. The method of claim 2, wherein the positive samples are 500 and the negative samples are 5000.
4. The method of claim 1, wherein the multi-scale feature extraction is performed in the search area updated in the previous frame by using a feature extraction network to obtain the matching template feature and the search area feature, and the specific method is as follows:
The matched template is subjected to convolutional neural network to obtain 6 multiplied by 256 matched template characteristics; the search area is subjected to convolutional neural network to obtain 22×22×256 search area characteristics.
5. The method of claim 4, wherein the RPN network performs a matching operation according to the matching template feature and the search area feature to obtain a matching tracking result, and the specific method is as follows:
Matching the template features and the search area features, and inputting the matched template features and the search area features into classification branches of the RPN network to obtain a classification response chart;
matching the template features and the search area features, and inputting the matched template features and the search area features into regression branches of the RPN network to obtain a regression response graph;
and adjusting the position of the target according to the classification response graph and the regression response graph to obtain a matching tracking result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111527248.XA CN114332157B (en) | 2021-12-14 | 2021-12-14 | Long-time tracking method for double-threshold control |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111527248.XA CN114332157B (en) | 2021-12-14 | 2021-12-14 | Long-time tracking method for double-threshold control |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114332157A CN114332157A (en) | 2022-04-12 |
| CN114332157B true CN114332157B (en) | 2024-05-24 |
Family
ID=81050127
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111527248.XA Active CN114332157B (en) | 2021-12-14 | 2021-12-14 | Long-time tracking method for double-threshold control |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114332157B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114972435A (en) * | 2022-06-10 | 2022-08-30 | 东南大学 | Target tracking method based on long-time and short-time integrated appearance updating mechanism |
| CN114782495B (en) * | 2022-06-16 | 2022-10-18 | 西安中科立德红外科技有限公司 | Multi-target tracking method, system and computer storage medium |
| CN117078976B (en) * | 2023-10-16 | 2024-01-30 | 华南师范大学 | Action scoring method, action scoring device, computer equipment and storage medium |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110443827A (en) * | 2019-07-22 | 2019-11-12 | 浙江大学 | A kind of UAV Video single goal long-term follow method based on the twin network of improvement |
| CN111144364A (en) * | 2019-12-31 | 2020-05-12 | 北京理工大学重庆创新中心 | Twin network target tracking method based on channel attention updating mechanism |
| CN111639551A (en) * | 2020-05-12 | 2020-09-08 | 华中科技大学 | Online multi-target tracking method and system based on twin network and long-short term clues |
| CN112132856A (en) * | 2020-09-30 | 2020-12-25 | 北京工业大学 | Twin network tracking method based on self-adaptive template updating |
| CN113052874A (en) * | 2021-03-18 | 2021-06-29 | 上海商汤智能科技有限公司 | Target tracking method and device, electronic equipment and storage medium |
| CN113591592A (en) * | 2021-07-05 | 2021-11-02 | 珠海云洲智能科技股份有限公司 | Overwater target identification method and device, terminal equipment and storage medium |
| WO2021227519A1 (en) * | 2020-05-15 | 2021-11-18 | 深圳市优必选科技股份有限公司 | Target tracking method and apparatus, and computer-readable storage medium and robot |
-
2021
- 2021-12-14 CN CN202111527248.XA patent/CN114332157B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110443827A (en) * | 2019-07-22 | 2019-11-12 | 浙江大学 | A kind of UAV Video single goal long-term follow method based on the twin network of improvement |
| CN111144364A (en) * | 2019-12-31 | 2020-05-12 | 北京理工大学重庆创新中心 | Twin network target tracking method based on channel attention updating mechanism |
| CN111639551A (en) * | 2020-05-12 | 2020-09-08 | 华中科技大学 | Online multi-target tracking method and system based on twin network and long-short term clues |
| WO2021227519A1 (en) * | 2020-05-15 | 2021-11-18 | 深圳市优必选科技股份有限公司 | Target tracking method and apparatus, and computer-readable storage medium and robot |
| CN112132856A (en) * | 2020-09-30 | 2020-12-25 | 北京工业大学 | Twin network tracking method based on self-adaptive template updating |
| CN113052874A (en) * | 2021-03-18 | 2021-06-29 | 上海商汤智能科技有限公司 | Target tracking method and device, electronic equipment and storage medium |
| CN113591592A (en) * | 2021-07-05 | 2021-11-02 | 珠海云洲智能科技股份有限公司 | Overwater target identification method and device, terminal equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114332157A (en) | 2022-04-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114332157B (en) | Long-time tracking method for double-threshold control | |
| CN108470332B (en) | Multi-target tracking method and device | |
| EP4044117A1 (en) | Target tracking method and apparatus, electronic device, and computer-readable storage medium | |
| CN109800689B (en) | Target tracking method based on space-time feature fusion learning | |
| WO2020215492A1 (en) | Multi-bernoulli multi-target video detection and tracking method employing yolov3 | |
| Chen et al. | Convolutional regression for visual tracking | |
| CN112668483B (en) | Single-target person tracking method integrating pedestrian re-identification and face detection | |
| WO2017015947A1 (en) | A system and a method for object tracking | |
| KR102132722B1 (en) | Tracking method and system multi-object in video | |
| CN114241548A (en) | Small target detection algorithm based on improved YOLOv5 | |
| CN109190544B (en) | Human identity recognition method based on sequence depth image | |
| Zulkifley | Two streams multiple-model object tracker for thermal infrared video | |
| CN112434599B (en) | Pedestrian re-identification method based on random occlusion recovery of noise channel | |
| CN111160407A (en) | Deep learning target detection method and system | |
| CN107844739B (en) | Robust target tracking method based on self-adaptive simultaneous sparse representation | |
| CN111192294A (en) | Target tracking method and system based on target detection | |
| CN108846850B (en) | Target tracking method based on TLD algorithm | |
| CN113963026A (en) | Target tracking method and system based on non-local feature fusion and online updating | |
| CN112446900A (en) | Twin neural network target tracking method and system | |
| CN110648351B (en) | Multi-appearance model fusion target tracking method and device based on sparse representation | |
| CN109272036B (en) | Random fern target tracking method based on depth residual error network | |
| Yang et al. | Online video text detection with markov decision process | |
| CN113888603A (en) | Loop detection and visual SLAM method based on optical flow tracking and feature matching | |
| CN110956649A (en) | Method and device for tracking multi-target three-dimensional object | |
| CN110781803B (en) | Human body posture recognition method based on extended 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |