CN115937260A - Multi-target tracking method, system, equipment and medium for joint measurement of central point offset and GIoU distance - Google Patents
Multi-target tracking method, system, equipment and medium for joint measurement of central point offset and GIoU distance Download PDFInfo
- Publication number
- CN115937260A CN115937260A CN202310014539.1A CN202310014539A CN115937260A CN 115937260 A CN115937260 A CN 115937260A CN 202310014539 A CN202310014539 A CN 202310014539A CN 115937260 A CN115937260 A CN 115937260A
- Authority
- CN
- China
- Prior art keywords
- tracking
- track
- detection frame
- giou
- target
- 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
Images
Landscapes
- Image Analysis (AREA)
Abstract
A multi-target tracking method, a system, equipment and a medium for central point offset and GIoU distance combined measurement are disclosed, firstly, the multi-target tracking method is divided into a high-branch detection frame and a low-branch detection frame according to confidence degree scores of the detection frames, secondly, hungary matching is carried out on the high-branch detection frame and a tracking track, then, greedy matching is carried out on the tracking track which is not matched with the low-branch detection frame, a tracking track is newly established for the high-branch detection frame which is not matched with the upper tracking track, 15-45 frames of the tracking track which is not matched with the upper detection frame are kept in a track pool as optimization, 30 frames are kept, matching is carried out when the tracking track appears again, and if the upper detection frame is not matched again in 30 frames, the tracking track is deleted in the track pool; the system, the device and the medium can detect the tracking target, store the computer program and execute the tracking method; according to the invention, the high-score detection frame and the low-score detection frame are matched twice, so that the target loss probability is reduced, the precision of multi-target tracking identification and the multi-target tracking accuracy are improved, and the target tracking effect is improved.
Description
Technical Field
The invention relates to the technical field of target tracking, in particular to a multi-target tracking method, a multi-target tracking system, multi-target tracking equipment and a multi-target tracking medium, wherein the multi-target tracking method is used for measuring central point offset and a GIoU distance in a combined mode.
Background
Multi-target tracking, which is a medium-level task of computer vision, is receiving more and more attention due to its academic value and commercial potential, and has important application significance and research value, is aimed at obtaining their respective motion trajectories by analyzing given videos to detect and track objects belonging to one or more categories, on the premise that the appearance and number of the objects of interest are unknown. In most of the existing multi-target tracking methods, threshold screening of detection confidence is carried out on a target during correlation, and the identity of the target is obtained through a detection frame with a correlation score higher than a certain threshold; for an object with a low detection score, for example, an object shielded by a large area, because the association score of the object is lower than a threshold value, the object can be ignored during tracking, which results in the loss of a tracking target and reduces the multi-target tracking effect. Although most detection frames are used in the BYTE method, the multi-target tracking accuracy is improved, and the high-order tracking precision of multi-target tracking is reduced.
The patent number CN202110130055.4 discloses a training method of a multi-target tracking model and a multi-target tracking method, wherein the method comprises the steps of constructing a target graph according to a target to be tracked in a current video frame; the top point of the target graph corresponds to a target to be tracked, and graph matching is carried out on the target graph and the existing track graph so as to calculate the matching score between the target to be tracked and the tracked track in the track graph; the top point of the track graph corresponds to the existing tracked track, and the matching track of the target to be tracked is determined according to the matching score; however, the method has the problems of reduced tracking precision and overlarge equipment load due to repeated and complicated steps and large calculation amount.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention aims to provide a multi-target tracking method, a multi-target tracking system, a multi-target tracking device and a multi-target tracking medium for measuring the central point offset and the GIoU distance in a combined manner.
In order to achieve the purpose, the invention adopts the technical scheme that:
a multi-target tracking method for central point offset and GIoU distance combined measurement comprises the following steps:
the method comprises the following steps: dividing the detection frames obtained by each detection network into high-resolution detection frames and low-resolution detection frames according to the confidence score of the detection frames, totally performing matching twice, separately processing the high-resolution frames and the low-resolution frames, excavating real objects from the low-resolution frames, and filtering out the background;
step two: performing Hungarian matching on the first matching by using a high score detection box and a previous tracking track, wherein a similarity matrix C h,tf As shown in formula 1):
C h,tf ＝dist 1 +1-GIoU
wherein the subscript tf is a track for first matching, h is a high-resolution detection box, in the formula of GIoU and IoU, A is the high-resolution detection box, B is the track box, C is the minimum box capable of enclosing the rectangular boxes A and B, dist 1 Is a two-dimensional tracking offset between the track obtained after the GIoU constraint and the detection frame; setting the tracking offset of the detection frame and the track frame to infinity when the GIoU value is smaller than the set threshold, as shown in equation 2):
step three: and performing greedy matching on the low-frequency detection frame and the tracking track which is not matched with the high-frequency detection frame for the first time again, wherein the similarity matrix of the second matching is as shown in the formula 3):
C l,ts ＝dist 2 +1-GIoU 3)
wherein the subscript ts is the track for the second matching, l is the low score detection box, the same as the first matching, dist 2 The two-dimensional tracking offset between the track obtained after the GIoU constraint and the low-resolution detection frame is used, the difference from the first matching is that the GIoU threshold value is generally set to be larger, and the threshold value can be selected for screening according to the used online multi-target tracking method;
step four: initializing the new track: building a tracking track for a detection frame which has no matching upper tracking track and has a higher score; delete or keep the old track: for the tracking track which does not match the upper detection frame, the tracking track is kept for 15-45 frames in the track pool, preferably, 30 frames are kept and are matched when the tracking track appears again, and if the tracking track does not match the upper detection frame again in 30 frames, the tracking track is deleted in the track pool.
A multi-target tracking system for central point offset and GIoU distance joint measurement comprises:
the detection module is used for detecting the tracking target and obtaining a detection frame of the tracking target and the confidence coefficient of the detection frame;
the association matching module is used for dividing the detection frame obtained by the detection module into a high-score detection frame and a low-score detection frame according to the confidence coefficient, and performing track matching on the high-score detection frame and the low-score detection frame twice so as to recover the track of the shielded object and reduce the loss of the tracked target and the track;
and the video result storage module is used for storing the result of each frame of the input video stream into a video file after the detection is finished and displaying the tracking effect.
A multi-target tracking device for joint measurement of center point offset and GIoU distance comprises:
a memory for storing a computer program;
the processor is used for realizing a multi-target tracking method of central point offset and GIoU distance combined measurement when the computer program is executed;
and the detector is used for detecting the tracking target, and various detectors can be selected.
A computer-readable storage medium storing a computer program, the computer program characterized in that: the computer program when executed by the processor enables tracking of multiple targets based on the midpoint offset and GIoU distance joint metrics.
Compared with the prior art, the invention has the beneficial effects that:
according to the invention, the high-score detection frame and the low-score detection frame are matched twice, so that the target loss probability is reduced, the multi-target tracking identification precision is improved, the multi-target tracking accuracy is improved, and the target tracking effect is improved.
Drawings
FIG. 1 is a flow chart of the present invention.
FIG. 2 shows the effect of the present invention.
Detailed Description
The structural and operational principles of the present invention are described in detail below with reference to the accompanying drawings.
As shown in fig. 1, a multi-target tracking method for central point offset and GIoU distance joint measurement includes the following steps:
the method comprises the following steps: dividing the detection frames obtained by each detection network into high-resolution detection frames and low-resolution detection frames according to the confidence score of the detection frames, totally performing matching twice, separately processing the high-resolution frames and the low-resolution frames, excavating real objects from the low-resolution frames, and filtering out the background;
step two: performing Hungarian matching on the first matching by using a high-score detection box and previous tracking tracks, wherein a similarity matrix C h,tf As shown in formula 1):
C h,tf ＝dist 1 +1-GIoU
in the formula of GIoU and IoU, A is the high-resolution detection frame, B is the track frame, C is the minimum frame capable of surrounding the rectangular frames A and B, dist 1 Is a two-dimensional tracking offset between the track obtained after the GIoU constraint and the detection frame; for the tracking offset of the detection box and the track box, the GIoU value less than the set threshold is set to infinity, e.g.
Formula 2):
step three: and performing greedy matching on the low-frequency detection frame and the tracking track which is not matched with the high-frequency detection frame for the first time again, wherein the similarity matrix of the second matching is as shown in the formula 3):
C l,ts ＝dist 2 +1-GIoU 3)
wherein the subscript ts is the track for the second matching, l is the low score detection box, the same as the first matching, dist 2 The two-dimensional tracking offset between the track obtained after the GIoU constraint and the low-resolution detection frame is used, the difference from the first matching is that the GIoU threshold value is generally set to be larger, and the threshold value can be selected for screening according to the used online multi-target tracking method;
step four: initializing the new track: building a tracking track for a detection frame which has no matching upper tracking track and has a higher score; delete or keep the old track: and keeping 15-45 frames of the tracking track which is not matched with the upper detection frame in a track pool, preferably, keeping 30 frames, matching when the tracking track appears again, and deleting the tracking track in the track pool if the tracking track is not matched with the upper detection frame again in 30 frames.
A multi-target tracking system for central point offset and GIoU distance joint measurement comprises:
the detection module is used for detecting the tracking target and obtaining a detection frame of the tracking target and the confidence of the detection frame;
the association matching module is used for dividing the detection frame obtained by the detection module into a high-score detection frame and a low-score detection frame according to the confidence coefficient, and performing track matching on the high-score detection frame and the low-score detection frame twice so as to recover the track of the shielded object and reduce the loss of the tracked target and the track;
and the video result storage module is used for storing each frame of result of the input video stream into a video file after detection is finished and displaying the tracking effect.
A multi-target tracking device for joint measurement of center point offset and GIoU distance comprises:
a memory for storing a computer program;
the processor is used for realizing a multi-target tracking method of the central point offset and the GIoU distance joint measurement when the computer program is executed;
and the detector is used for detecting the tracking target, and various detectors can be selected.
A computer-readable storage medium storing a computer program, characterized in that: the computer program, when executed by a processor, is capable of tracking multiple targets based on a combined metric of midpoint offset and GIoU distance.
As shown in fig. 2, which is an effect diagram of the present invention, different colored boxes in the diagram represent different objects, and the numbers in the upper left corner of the boxes represent the IDs of the objects. Compared with the prior art, the experiment shows that the high-order tracking precision of multi-target tracking identification is improved by 0.67%, the multi-target tracking accuracy is improved by 0.2%, the accuracy of F value identification is improved by 1.8%, and the ID switching frequency is reduced by 33.5%.
Claims (4)
1. A multi-target tracking method for central point offset and GIoU distance combined measurement is characterized by comprising the following steps:
the method comprises the following steps: dividing the detection frames obtained by each detection network into a high-frequency detection frame and a low-frequency detection frame according to the confidence score of the detection frames, totally performing matching twice, separately processing the high-frequency detection frame and the low-frequency detection frame, excavating a real object from the low-frequency detection frame, and filtering a background;
step two: performing Hungarian matching on the first matching by using a high score detection box and a previous tracking track, wherein a similarity matrix C h,tf As shown in formula 1):
in the formula of GIoU and IoU, A is the high-resolution detection frame, B is the track frame, C is the minimum frame capable of surrounding the rectangular frames A and B, dist 1 The two-dimensional tracking offset between the track and the detection frame is obtained after GIoU constraint; setting the tracking offset of the detection frame and the track frame to infinity when the GIoU value is smaller than the set threshold, as shown in equation 2):
step three: and carrying out greedy matching on the low-resolution detection frame and the tracking track which is not matched with the high-resolution detection frame for the first time again, wherein the similarity matrix of the second matching is as shown in formula 3):
C l,ts ＝dist 2 +1-GIoU 3)
wherein the subscript ts is the track for the second matching, l is the low score detection box, the same as the first matching, dist 2 The two-dimensional tracking offset between the track obtained after the GIoU constraint and the low-resolution detection frame is used, the difference from the first matching is that the GIoU threshold value is generally set to be larger, and the threshold value can be selected for screening according to the used online multi-target tracking method;
step four: initializing the new track: building a tracking track for a detection frame which has no matching upper tracking track and has a higher score; delete or keep the old track: for the tracking track which does not match the upper detection frame, the tracking track is kept for 15-45 frames in the track pool, preferably, 30 frames are kept and are matched when the tracking track appears again, and if the tracking track does not match the upper detection frame again in 30 frames, the tracking track is deleted in the track pool.
2. A multi-target tracking system for central point offset and GIoU distance combined measurement is characterized in that: the method comprises the following steps:
the detection module is used for detecting the tracking target and obtaining a detection frame of the tracking target and the confidence of the detection frame;
the association matching module is used for dividing the detection frame obtained by the detection module into a high-score detection frame and a low-score detection frame according to the confidence coefficient, and performing track matching on the high-score detection frame and the low-score detection frame twice so as to recover the track of the shielded object and reduce the loss of the tracked target and the track;
and the video result storage module is used for storing each frame of result of the input video stream into a video file after detection is finished and displaying the tracking effect.
3. A multi-target tracking device for central point offset and GIoU distance joint measurement is characterized in that: the method comprises the following steps:
a memory for storing a computer program;
the processor is used for realizing a multi-target tracking method of the central point offset and the GIoU distance joint measurement when the computer program is executed;
and the detector is used for detecting the tracking target, and various detectors can be selected.
4. A computer-readable storage medium storing a computer program, the computer program characterized in that: the computer program, when executed by a processor, is capable of tracking multiple targets based on a combined metric of midpoint offset and GIoU distance.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310014539.1A CN115937260A (en) | 2023-01-05 | 2023-01-05 | Multi-target tracking method, system, equipment and medium for joint measurement of central point offset and GIoU distance |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310014539.1A CN115937260A (en) | 2023-01-05 | 2023-01-05 | Multi-target tracking method, system, equipment and medium for joint measurement of central point offset and GIoU distance |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115937260A true CN115937260A (en) | 2023-04-07 |
Family
ID=86649145
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310014539.1A Pending CN115937260A (en) | 2023-01-05 | 2023-01-05 | Multi-target tracking method, system, equipment and medium for joint measurement of central point offset and GIoU distance |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115937260A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116703983A (en) * | 2023-06-14 | 2023-09-05 | 石家庄铁道大学 | Combined shielding target detection and target tracking method |
-
2023
- 2023-01-05 CN CN202310014539.1A patent/CN115937260A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116703983A (en) * | 2023-06-14 | 2023-09-05 | 石家庄铁道大学 | Combined shielding target detection and target tracking method |
| CN116703983B (en) * | 2023-06-14 | 2023-12-19 | 石家庄铁道大学 | Combined shielding target detection and target tracking method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105405151B (en) | Anti-shelter target tracking based on particle filter and weighting Surf | |
| CN111080673B (en) | Anti-occlusion target tracking method | |
| CN109544592B (en) | Moving object detection algorithm for camera movement | |
| CN110490907B (en) | Moving target tracking method based on multi-target feature and improved correlation filter | |
| CN104574439A (en) | Kalman filtering and TLD (tracking-learning-detection) algorithm integrated target tracking method | |
| CN101383899A (en) | Video image stabilizing method for space based platform hovering | |
| CN112488057A (en) | Single-camera multi-target tracking method utilizing human head point positioning and joint point information | |
| CN102915545A (en) | OpenCV(open source computer vision library)-based video target tracking algorithm | |
| CN109239702B (en) | Airport low-altitude flying bird number statistical method based on target state set | |
| CN113256690B (en) | Pedestrian multi-target tracking method based on video monitoring | |
| CN114092517A (en) | Multi-target tracking method based on traditional and deep learning algorithm | |
| CN111161325A (en) | Three-dimensional multi-target tracking method based on Kalman filtering and LSTM | |
| CN107808524A (en) | A kind of intersection vehicle checking method based on unmanned plane | |
| CN115937260A (en) | Multi-target tracking method, system, equipment and medium for joint measurement of central point offset and GIoU distance | |
| CN103729462B (en) | A kind of pedestrian retrieval method blocked based on rarefaction representation process | |
| CN111402303A (en) | Target tracking architecture based on KFSTRCF | |
| CN114926859A (en) | Pedestrian multi-target tracking method in dense scene combined with head tracking | |
| CN115984320A (en) | Multi-target tracking method based on long-short-term motion prediction and ordered data association | |
| CN110443142A (en) | A kind of deep learning vehicle count method extracted based on road surface with segmentation | |
| CN103996207A (en) | Object tracking method | |
| CN114550219B (en) | Pedestrian tracking method and device | |
| Roopchand et al. | Bat detection and tracking toward batsman stroke recognition | |
| CN110889347A (en) | Density traffic flow counting method and system based on space-time counting characteristics | |
| CN107067411B (en) | Mean-shift tracking method combined with dense features | |
| CN112561956B (en) | Video target tracking method and device, electronic equipment and storage medium |
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 |