Different target detection and switching tracking method based on deep learning
Through deep learning-based methods to detect and switch targets, the problem that the existing technology cannot automatically switch and track different targets is solved, and efficient and automated target switching tracking is achieved.
Patent Information
- Application Number
- CN202411936241.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-26
- Publication Date
- 2025-05-09
AI Technical Summary
Existing target tracking methods cannot automatically switch to tracking for another type of target.
Using a deep learning-based method, the key detection area is set by detecting the longitudinal coordinates, lateral coordinates and target width of the first type of target, the second type of target is detected in this area using the deep learning network model, and whether to switch tracking is determined based on preset conditions.
Automatic switching tracking between two types of targets is achieved, improving the degree of automation, real-time and anti-interference ability.
Smart Images
Figure CN119963598A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of target tracking technology, and in particular to a method for detecting and switching different targets based on deep learning. Background Art
[0002] Target tracking means that after detecting the target in the image, the target is numbered according to the continuous images, so that the same target can be continuously located in the continuous images. Target tracking can improve the accuracy of subsequent target detection and continuously track the target's motion state. It is widely used in the fields of unmanned driving, security monitoring, and behavior recognition.
[0003] However, current target tracking methods can only track one type of target and cannot automatically switch to tracking another type of target. Summary of the invention
[0004] The present invention aims to provide a method for detecting and switching different targets based on deep learning, so as to solve the problem of switching tracking between two types of targets.
[0005] To achieve the above object, the present invention adopts the following technical solution:
[0006] On the one hand, this specification provides a method for detecting and tracking different targets based on deep learning, including:
[0007] S1, automatically track the first type of target in the field of view according to the target tracking instruction;
[0008] S2. When the first type of target throws out the second type of target, a key detection area is set according to the longitudinal coordinate, the transverse coordinate and the target width of the first type of target, and the second type of target is detected in the key detection area based on the deep learning network model;
[0009] S3, recording the second-category target detected in the key detection area as the second-category target to be switched and tracked, and obtaining the longitudinal coordinate and the transverse coordinate of the second-category target to be switched and tracked;
[0010] S4, judging whether the second type of target to be switched to track meets a preset tracking switching condition according to the longitudinal coordinate and the transverse coordinate of the first type of target and the longitudinal coordinate and the transverse coordinate of the second type of target to be switched to track;
[0011] S5. If the preset tracking switching condition is met, the tracking target is switched to the second type of target to be switched.
[0012] On the other hand, this specification also provides a different target detection and switching tracking device based on deep learning, including:
[0013] A first-class target tracking module, used for automatically tracking a first-class target in the field of view according to a target tracking instruction;
[0014] A second-category target detection module, configured to, when the first-category target throws out the second-category target, set a key detection area according to the longitudinal coordinates, transverse coordinates and target width of the first-category target, and detect the second-category target in the key detection area based on a deep learning network model;
[0015] A second-category target acquisition module, used to record the second-category target detected in the key detection area as the second-category target to be switched and tracked, and to acquire the longitudinal coordinate and the transverse coordinate of the second-category target to be switched and tracked;
[0016] A switching condition judgment module, used to judge whether the second type of target to be switched to track meets a preset tracking switching condition according to the longitudinal coordinate and the transverse coordinate of the first type of target and the longitudinal coordinate and the transverse coordinate of the second type of target to be switched to track;
[0017] The tracking switching module is used to switch the tracking target to the second type of target to be switched if the preset tracking switching condition is met.
[0018] On the other hand, this specification also provides an electronic device,
[0019] A processor; and a memory arranged to store computer executable instructions, which when executed cause the processor to perform the steps of the method as claimed in any one of the preceding claims.
[0020] Based on the above technical solution, this specification can achieve the following technical effects:
[0021] The present invention proposes a method for detecting and switching different targets based on deep learning: according to the target tracking instruction, the first type of target in the field of view is automatically tracked; when the first type of target throws out the second type of target, the key detection area is set according to the longitudinal coordinate, transverse coordinate and target width of the first type of target, and the second type of target is detected in the key detection area based on the deep learning network model; the second type of target detected in the key detection area is recorded as the second type of target to be switched and tracked, and the longitudinal coordinate and transverse coordinate of the second type of target to be switched and tracked are obtained; according to the longitudinal coordinate and transverse coordinate of the first type of target and the longitudinal coordinate and transverse coordinate of the second type of target to be switched and tracked, it is judged whether the second type of target to be switched and tracked meets the preset tracking switching condition; if the preset tracking switching condition is met, the tracking target is switched to the second type of target to be switched and tracked. That is, according to the trajectory and spatial distribution relationship between the first type of target and the second type of target, the second type of target is detected and identified, and the pseudo target interference is eliminated according to the distribution characteristics of the position of the second type of target in the field of view between frames, so as to realize the automatic switching tracking of the two types of targets. The method has the characteristics of high automation, good real-time performance, high reliability of target switching tracking, and anti-interference. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] Figure 1 It is a flowchart of a method for detecting and switching different targets based on deep learning provided in Example 1 of this specification;
[0023] Figure 2 This is a schematic diagram of a different target detection and switching tracking device based on deep learning provided in Example 3 of this specification. DETAILED DESCRIPTION
[0024] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The advantages and features of the present invention will become more apparent from the following description and claims. It should be noted that the drawings are all in very simplified form and are not in precise proportions, and are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present invention.
[0025] It should be noted that, in order to clearly explain the content of the present invention, the present invention specifically cites multiple embodiments to further illustrate different implementations of the present invention, wherein the multiple embodiments are enumerated rather than exhaustive. In addition, for the sake of brevity of explanation, the contents mentioned in the previous embodiments are often omitted in the subsequent embodiments. Therefore, the contents not mentioned in the subsequent embodiments can refer to the previous embodiments accordingly.
[0026] Example 1
[0027] Please refer to Figure 1 , Figure 1The figure shows a flow chart of a method for detecting and switching different targets based on deep learning provided by this embodiment. This method is widely used in a wide range of scenarios. For example, in a rescue scenario, after an aircraft drops supplies, the aircraft is recorded as a first-class target and the supplies are recorded as a second-class target. The aircraft target and the supply target can be automatically switched to be tracked.
[0028] The method specifically comprises the following steps:
[0029] S1, automatically track the first type of target in the field of view according to the target tracking instruction;
[0030] It should be noted that one implementation of S1 may be:
[0031] Through the detection method based on deep learning, the network model used is the single-stage detector YOLOv5, which simultaneously detects and identifies various types of targets in the field of view and establishes a target pool. According to the tracking instructions sent by the operator, the first type of target in the field of view is automatically tracked. If the first type of target is not identified, no action is required.
[0032] S2. When the first type of target throws out the second type of target, a key detection area is set according to the longitudinal coordinate, the transverse coordinate and the target width of the first type of target, and the second type of target is detected in the key detection area based on the deep learning network model;
[0033] It should be noted that one implementation of S2 may be:
[0034] When the first type of target throws out the second type of target, assuming the longitudinal coordinate of the first type of target is Y, the transverse coordinate is X and the target width is a, the key detection area is set to a pixel area of (X+a)×(Y+100).
[0035] The single-stage detector YOLOv5 is used to detect the second type of target in the pixel area of (X+a)×(Y+100).
[0036] S3, recording the second-category target detected in the key detection area as the second-category target to be switched and tracked, and obtaining the longitudinal coordinate and the transverse coordinate of the second-category target to be switched and tracked;
[0037] It should be noted that one implementation of S3 may be:
[0038] The second type of target detected in the key detection area is recorded as the second type of target to be switched and tracked;
[0039] Furthermore, the second type of target to be switched and tracked is continuously detected to determine whether the second type of target to be switched and tracked is stably detected for N consecutive frames; if the second type of target to be switched and tracked is stably detected for N consecutive frames, the longitudinal coordinate and the lateral coordinate of the second type of target to be switched and tracked are obtained. If the second type of target to be switched and tracked is not stably detected for N consecutive frames, the second type of target to be switched and tracked is eliminated.
[0040] Based on this, the reliability and anti-interference of target switching tracking are improved.
[0041] S4, judging whether the second type of target to be switched to track meets a preset tracking switching condition according to the longitudinal coordinate and the transverse coordinate of the first type of target and the longitudinal coordinate and the transverse coordinate of the second type of target to be switched to track;
[0042] It should be noted that one implementation of S4 may be:
[0043] According to the longitudinal coordinates and transverse coordinates of the first type of target and the longitudinal coordinates and transverse coordinates of the second type of target to be switched for tracking, the longitudinal distance and transverse distance between the first type of target and the second type of target to be switched for tracking are calculated; according to the longitudinal distance and transverse distance between the first type of target and the second type of target to be switched for tracking, whether the second type of target to be switched for tracking meets the preset tracking switching condition is judged, specifically:
[0044] Determine whether the longitudinal distance and the lateral distance between the first type of target and the second type of target to be switched and tracked meet the following conditions at the same time:
[0045] 1. The longitudinal distance between the first type of target and the second type of target to be switched to track continues to increase;
[0046] 2. The lateral distance between the first type of target and the second type of target to be switched to track does not exceed the target width of the first type of target;
[0047] If the above conditions are met at the same time, the second type of target to be switched to track meets the preset tracking switching condition; if any one of them is not met, the second type of target to be switched to track does not meet the preset tracking switching condition.
[0048] S5. If the preset tracking switching condition is met, the tracking target is switched to the second type of target to be switched.
[0049] It should be noted that one implementation of S5 may be:
[0050] If the preset tracking switching condition is met, the second category target to be switched to track is identified as the second category target, and the system tracking target is switched to the second category target.
[0051] Furthermore, this embodiment also provides a method for removing false interference:
[0052] Obtain the longitudinal coordinates and transverse coordinates of the second type of target in the switching tracking in the current frame and the previous frame in the field of view; and determine whether the second type of target in the switching tracking has reverse motion in the longitudinal position or the transverse position based on the longitudinal coordinates and the transverse coordinates of the current frame and the previous frame.
[0053] If yes, the second type of target to be tracked is recorded as a false target and removed. At the same time, the tracking of the second type of target in this frame image fails, and the next frame tracking procedure is entered. If no, the second type of target is continuously tracked until the end.
[0054] In summary, the present invention proposes a method for detecting and switching different targets based on deep learning: according to the target tracking instruction, automatically track the first type of target in the field of view; when the first type of target throws out the second type of target, set the key detection area according to the longitudinal coordinate, transverse coordinate and target width of the first type of target, and detect the second type of target in the key detection area based on the deep learning network model; record the second type of target detected in the key detection area as the second type of target to be switched and tracked, and obtain the longitudinal coordinate and transverse coordinate of the second type of target to be switched and tracked; according to the longitudinal coordinate and transverse coordinate of the first type of target and the longitudinal coordinate and transverse coordinate of the second type of target to be switched and tracked, judge whether the second type of target to be switched and tracked meets the preset tracking switching condition; if the preset tracking switching condition is met, switch the tracking target to the second type of target to be switched and tracked. That is, according to the trajectory and spatial distribution relationship between the first type of target and the second type of target, detect and identify the second type of target, according to the distribution characteristics of the position of the second type of target in the field of view between frames, eliminate the interference of pseudo targets, and realize the automatic switching tracking of the two types of targets. This method has the characteristics of high automation, good real-time performance, high reliability of target switching tracking, and anti-interference.
[0055] Example 2
[0056] This embodiment provides another specific implementation of a different target detection and switching tracking method based on deep learning.
[0057] The steps include:
[0058] Step 1: Through the deep learning-based detection method, the network model used is the single-stage detector YOLOv5, which simultaneously detects and identifies various targets in the field of view and establishes a target pool;
[0059] Step 2: According to the tracking command sent by the operator, the first type of target in the field of view is automatically tracked. If the first type of target is not identified, no action is required;
[0060] Step 3: Detect and identify the second type of targets based on the trajectory and spatial distribution relationship between the first type of targets and the second type of targets:
[0061] The pixel area of a×100 (a is the width of the first type of target) in the longitudinal direction of the first type of target is used as the key detection area of the second type of target. When the first type of target throws out the second type of target, the second type of target detected in the key detection area is used as the second type of target to be switched and tracked, and a target chain continuous detection is established;
[0062] Step 4: stably detect the second-type target for more than 10 frames, calculate the longitudinal distance and lateral distance between the second-type target and the first-type target, and if the longitudinal distance continues to increase and the lateral distance does not exceed the width of the first-type target, identify it as the second-type target, and switch the tracking target from the first-type target to the second-type target;
[0063] Step 5: Eliminate pseudo target interference based on the distribution characteristics of the second type of target's position between frames in the field of view:
[0064] Detect the progressive relationship between the position of the second type of target in the field of view and the position of the previous frame: If the longitudinal or lateral position moves in the opposite direction, it is determined that the tracking of the second type of target in this frame image fails, and the next frame tracking program is entered. Otherwise, the second type of target is continuously tracked until the end.
[0065] Example 3
[0066] Please refer to Figure 2 , Figure 2 The figure shows a schematic diagram of a different target detection and switching tracking device based on deep learning provided by this embodiment. The device includes:
[0067] A first-class target tracking module 302, used to automatically track a first-class target in the field of view according to a target tracking instruction;
[0068] A second-category target detection module 304, configured to, when the first-category target throws out the second-category target, set a key detection area according to the longitudinal coordinates, transverse coordinates and target width of the first-category target, and detect the second-category target in the key detection area based on a deep learning network model;
[0069] A second-category target acquisition module 306, configured to record the second-category target detected in the key detection area as the second-category target to be switched and tracked, and to acquire the longitudinal coordinate and the transverse coordinate of the second-category target to be switched and tracked;
[0070] A switching condition judgment module 308 is used to judge whether the second type of target to be switched to track meets a preset tracking switching condition according to the longitudinal coordinate and the transverse coordinate of the first type of target and the longitudinal coordinate and the transverse coordinate of the second type of target to be switched to track;
[0071] The tracking switching module 310 is used to switch the tracking target to the second type of target to be switched if a preset tracking switching condition is met.
[0072] Optionally, the device further includes a stability detection module, which is specifically used for:
[0073] Continuously detecting the second type of target to be switched and tracked, and determining whether the second type of target to be switched and tracked is stably detected for N consecutive frames;
[0074] If the second type of target to be switched and tracked is stably detected for N consecutive frames, the longitudinal coordinate and the lateral coordinate of the second type of target to be switched and tracked are obtained.
[0075] Based on this, the reliability of target switching tracking is improved.
[0076] Optionally, the device further includes an interference elimination module, which is specifically used for:
[0077] Obtain the longitudinal coordinates and the transverse coordinates of the second type of target in the current frame and the previous frame in the field of view;
[0078] According to the longitudinal coordinates and the transverse coordinates of the current frame and the previous frame, determining whether the second type of target to be tracked switches has reverse movement in the longitudinal position or the transverse position;
[0079] If so, the second type of target that is switched to be tracked is recorded as a false target and removed.
[0080] Based on this, the anti-interference performance of target switching tracking is further improved.
[0081] Example 4
[0082] In another feasible embodiment, this embodiment provides a device for different target detection and switching tracking based on deep learning, and the device may specifically include:
[0083] A processor; and a memory arranged to store computer executable instructions, which, when executed, cause the processor to perform the steps in any of the above method embodiments.
[0084] Example 5
[0085] In another feasible embodiment, this embodiment provides a storage medium for different target detection and switching tracking based on deep learning, and the storage medium may specifically include:
[0086] The storage medium stores a processing program for detecting and switching tracks of different targets based on deep learning, and when the processing program for detecting and switching tracks of different targets based on deep learning is executed by the processor, the steps in any of the above method embodiments are implemented.
[0087] The above description is only an embodiment of the present specification and is not intended to limit the present specification. For those skilled in the art, the present specification can be modified and varied in various ways. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present specification should be included in the scope of the claims of the present specification.
Claims
1. A method for detecting and switching different targets based on deep learning, characterized in that: include: S1, automatically track the first type of target in the field of view according to the target tracking instruction; S2. When the first type of target throws out the second type of target, a key detection area is set according to the longitudinal coordinate, the transverse coordinate and the target width of the first type of target, and the second type of target is detected in the key detection area based on the deep learning network model; S3, recording the second-category target detected in the key detection area as the second-category target to be switched and tracked, and obtaining the longitudinal coordinate and the transverse coordinate of the second-category target to be switched and tracked; S4, judging whether the second type of target to be switched to track meets a preset tracking switching condition according to the longitudinal coordinate and the transverse coordinate of the first type of target and the longitudinal coordinate and the transverse coordinate of the second type of target to be switched to track; S5. If the preset tracking switching condition is met, the tracking target is switched to the second type of target to be switched.
2. The method according to claim 1, characterized in that The step of setting a key detection area according to the longitudinal coordinate, the transverse coordinate and the target width of the first type of target includes: Assume that the longitudinal coordinate of the first type of target is Y, the transverse coordinate is X and the target width is a; The key detection area is set to a pixel area of (X+a)×(Y+100).
3. The method according to claim 1, characterized in that The detecting the second type of target in the key detection area based on the deep learning network model includes: The single-stage detector YOLOv5 is used to detect the second type of target in the key detection area.
4. The method according to claim 1, characterized in that: After recording the second type of target detected in the key detection area as the second type of target to be switched and tracked, the method further includes: Continuously detecting the second type of target to be switched and tracked, and determining whether the second type of target to be switched and tracked is stably detected for N consecutive frames; If the second type of target to be switched and tracked is stably detected for N consecutive frames, the longitudinal coordinate and the lateral coordinate of the second type of target to be switched and tracked are obtained.
5. The method according to claim 4, characterized in that After continuously detecting the second type of target to be switched and tracked, and determining whether the second type of target to be switched and tracked is stably detected for N consecutive frames, the method further includes: If the second type of target to be switched and tracked cannot be stably detected for N consecutive frames, the second type of target to be switched and tracked is eliminated.
6. The method according to claim 1, characterized in that The step of judging whether the second type of target to be switched for tracking satisfies a preset tracking switching condition according to the longitudinal coordinate and the transverse coordinate of the first type of target and the longitudinal coordinate and the transverse coordinate of the second type of target to be switched for tracking comprises: Calculate the longitudinal distance and the lateral distance between the first type of target and the second type of target to be switched for tracking according to the longitudinal coordinate and the lateral coordinate of the first type of target and the longitudinal coordinate and the lateral coordinate of the second type of target to be switched for tracking; According to the longitudinal distance and the lateral distance between the first category target and the second category target to be switched for tracking, it is determined whether the second category target to be switched for tracking meets a preset tracking switching condition.
7. The method according to claim 6, characterized in that The determining, based on the longitudinal distance and the lateral distance between the first-category target and the second-category target to be switched for tracking, whether the second-category target to be switched for tracking meets a preset tracking switching condition includes: Determine whether the longitudinal distance and the lateral distance between the first type of target and the second type of target to be switched and tracked meet the following conditions at the same time: The longitudinal distance between the first type of target and the second type of target to be switched and tracked continues to increase; The lateral distance between the first type of target and the second type of target to be switched for tracking does not exceed the target width of the first type of target; If the above conditions are met at the same time, the second type of target to be switched to track meets the preset tracking switching conditions; If any one of the items is not met, the second type of target to be switched for tracking does not meet the preset tracking switching condition.
8. The method according to claim 1, characterized in that After the tracking target is switched to the second type of target to be switched and tracked, the method further includes: Obtain the longitudinal coordinates and the transverse coordinates of the second type of target in the current frame and the previous frame in the field of view; According to the longitudinal coordinates and the transverse coordinates of the current frame and the previous frame, determining whether the second type of target to be tracked switches has reverse movement in the longitudinal position or the transverse position; If so, the second type of target that is switched to be tracked is recorded as a false target and removed.
9. A different target detection and switching tracking device based on deep learning, characterized in that: include: A first-class target tracking module, used for automatically tracking a first-class target in the field of view according to a target tracking instruction; A second-category target detection module, configured to, when the first-category target throws out the second-category target, set a key detection area according to the longitudinal coordinates, transverse coordinates and target width of the first-category target, and detect the second-category target in the key detection area based on a deep learning network model; A second-category target acquisition module, used to record the second-category target detected in the key detection area as the second-category target to be switched and tracked, and to acquire the longitudinal coordinate and the transverse coordinate of the second-category target to be switched and tracked; A switching condition judgment module, used to judge whether the second type of target to be switched to track meets a preset tracking switching condition according to the longitudinal coordinate and the transverse coordinate of the first type of target and the longitudinal coordinate and the transverse coordinate of the second type of target to be switched to track; The tracking switching module is used to switch the tracking target to the second type of target to be switched if the preset tracking switching condition is met.
10. An electronic device, characterized in that: include: processor; and a memory arranged to store computer executable instructions which, when executed, cause the processor to perform the steps of the method as claimed in any one of claims 1 to 8.