Time-sharing detection tracking double-cycle method and device, system, storage medium

Abstract

This invention discloses a time-division detection and tracking dual-loop method, apparatus, system, and storage medium, comprising: obtaining a target capture pane and a topologically equivalent envelope model of the target region topology based on the initial frame image of a coupled temporal image of a UAV small target in consecutive frames; obtaining the envelope feature point set of the UAV small target in the initial frame and the extraction rules of the envelope feature points of consecutive frames based on the topologically equivalent envelope model of the target region; obtaining the predicted target capture pane for the current frame based on the envelope feature point set extracted from the initial frame and subsequent consecutive frames; performing inter-frame target envelope homotopy discrimination and ID consistency evaluation based on the extracted target envelope in the current frame pane and the target envelope of historical frames; and constructing an equally spaced time-division dual-loop mechanism based on detection and tracking to achieve ID calibration and stable tracking of inter-frame homotopy targets. The technical solution of this invention effectively reduces the false detection and false negative rates in UAV small target detection and improves the stability and low switching rate of ID in moving target tracking.

Description

TECHNICAL FIELD

[0001] The present application belongs to the target tracking technology in the field of computer vision detection, and particularly relates to a time-sharing detection and tracking double-cycle method, device, system and storage medium. BACKGROUND

[0002] With the rapid development of unmanned aerial vehicle technology, unmanned aerial vehicles have advantages of high mobility, strong concealment, low cost and lightness, and their applications in the fields of industry, agriculture, military and the like have shown explosive growth. From photography and videography, logistics transportation to environmental monitoring, disaster relief and even military reconnaissance, the wide application of unmanned aerial vehicles has created value, but also has derived the phenomena of "black flight" and "wild flight", which pose a great threat to personal privacy, air traffic and national security. Therefore, it is very important to enhance the monitoring and tracking capability of illegal intrusion unmanned aerial vehicles for maintaining the safety of the country and society.

[0003] Current unmanned aerial vehicle detection and tracking technologies mainly include radar, audio, radio and vision. The radar system realizes parameter solving by emitting electromagnetic waves and analyzing the Doppler frequency shift characteristics of the target echo, has the advantage of long-distance detection, but has limitations such as echo energy attenuation, susceptibility to electromagnetic interference and confusion with bird target characteristics in small target monitoring; the audio monitoring system is based on Mel frequency cepstrum coefficient (MFCC) and linear prediction cepstrum coefficient (LPCC) and other voiceprint feature extraction, and combines support vector machine (SVM) or deep learning algorithm for identification, has the advantages of low cost and easy deployment, but its effective range is seriously restricted by environmental noise and propagation attenuation, and is mostly used as an auxiliary means in actual application; the radio technology realizes target positioning by constructing an unmanned aerial vehicle communication protocol feature library and combining time difference positioning (TDOA) and direction finding cross positioning technology, can analyze flight state and control link information, but has a detection blind area for unmanned aerial vehicles using silent flight mode, and is susceptible to complex signal interference in the same frequency band; compared with other technologies, the vision detection and tracking technology captures video through a camera, realizes unmanned aerial vehicle target classification and spatial positioning based on image feature extraction, and has balanced performance in monitoring range, scene adaptability and detection accuracy, is especially suitable for real-time monitoring of low-altitude targets, has advantages of intuitive visualization, strong anti-interference and stable performance, and has become the mainstream technical solution in the field of low-altitude safety protection.

[0004] Although the vision detection and tracking technology shows significant advantages, it still faces challenges such as tracking trajectory interruption, target ID frequent jumping and significant decline in recognition and tracking accuracy when detecting and tracking small unmanned aerial vehicle targets with sparse features, degraded texture and complete occlusion in complex scenes. In addition, the existing vision tracking method usually adopts a mode of frame-by-frame identification and detection of collected video to realize target tracking, has the problems of large data processing amount and heavy computing load, and is difficult to meet the demand of real-time detection and tracking.

[0005] The Chinese patent with patent application number 202510332138.X and the invention name of "a UAV target tracking method". When the invention is implemented, first, the target boundary box and confidence are obtained based on the target detector, and the noise adaptive Kalman filter is used to realize the prediction of the target motion state; Then through the double judgment of the intersection-over-union matching degree and the confidence change rate, the target needing feature extraction is screened, and its appearance feature is extracted through the Rep-OSNet network; Then the adaptive cost function of fusing motion direction and appearance similarity is used for cascade matching, and the mismatched target and trajectory are verified again; Finally, the effective trajectory state is updated by using the noise adaptive Kalman filter, the long-term lost trajectory is deleted, and the target identity ID and the predicted trajectory box are output synchronously. Although this method reduces the computational overhead by using dynamic feature extraction strategy and lightweight network design, and improves the tracking robustness in the semi-occluded scene by using adaptive cost function, it is sensitive to target appearance deformation, so in the long-distance small target tracking scene, the tracking performance will be attenuated due to insufficient feature resolution.

[0006] The Chinese patent with patent application number 202410812107.X and the invention name of "single target tracking UAV based on improved TransT algorithm". The invention first divides the TIB-Net dataset with an area ratio of less than 1% of the UAV into training and test sets in the ratio of 7:3 and enhances the training set, and then constructs a TransT-drone UAV target tracking model. The model reduces the parameter amount by adding a high-resolution tracking branch and cropping the large target related layer, replaces Conv with SPD-Conv to enhance small target feature extraction, and introduces GAM attention mechanism in the multi-scale fusion stage to improve feature retention capability. Experiments show that the model precision, recall rate and mAP are improved by 11.9%, 15.2% and 9% respectively, and the parameter amount and model size are reduced by more than 50%. However, the strategy of cropping the large target related layer not only weakens the adaptability of the model to the target size, but also easily causes tracking box drift or loss, and causes the rupture of the associated high-level semantic feature chain, resulting in target ID jump and tracking interruption in complex scenes.

[0007] Patent application number 202311680555.0, invention name "a low altitude unmanned aerial vehicle target tracking method based on LA-DeepSORT". The invention first uses the enhanced correlation coefficient (ECC) algorithm to compensate for the motion of the unmanned aerial vehicle video, and obtains the video frame sequence by frame processing. Then, the pre-trained YOLOv5 model is used to detect the video frame, and the unmanned aerial vehicle target prediction box and confidence are obtained. According to these prediction boxes and confidence, the Kalman filter and the tracker introducing MobileNetV3 are initialized to enhance the appearance feature extraction capability. Finally, through Kalman filter prediction, Hungarian algorithm matching and parameter updating, the tracking results containing unmanned aerial vehicle ID and tracking trajectory information are output. Although the invention improves the tracking performance through multi-dimensional improvement, the detection ability of YOLOv5 for low altitude remote small targets is limited, and the frame processing of 30 frames per second may not meet the actual demand in higher frame rate or low delay scene.

[0008] The above three inventions all use the method of video frame separation step by step to realize the tracking of unmanned aerial vehicle targets. This leads to a sharp increase in the amount of data processing of the detection and tracking model, significantly reducing the model processing speed. At the same time, the above methods all rely on the texture features of unmanned aerial vehicle targets to different degrees. When the unmanned aerial vehicle is blocked by vegetation or buildings in complex environments, its tracking process may be interrupted, and after the shielding is removed, it needs to be detected again, resulting in a change of target ID. In addition, the first method is sensitive to target appearance deformation, the second method has weak adaptability to target size, and the third method is limited by the target recognition network, resulting in insufficient tracking ability of the model for remote small unmanned aerial vehicle targets. At the same time, the third method is limited by the frame processing rate, which makes the model unable to meet the demand of real-time tracking of unmanned aerial vehicle targets. Although the above inventions expand the unmanned aerial vehicle target tracking method, they are unable to realize real-time and continuous tracking of small targets in complex scenes. SUMMARY

[0009] To solve the problems existing in the prior art, the present application provides a kind of target envelope homotopy criterion's detection tracking double cycle method and device, system, storage medium.

[0010] To achieve the above purpose, the present application provides the following scheme: A kind of detection tracking double cycle method, characterized by comprising: real-time acquisition of long-short focus combined unmanned aerial vehicle small target continuous frame coupled time sequence image by ground-to-air image acquisition equipment; obtain target interception window and target region topological equivalent envelope model according to the initial frame image of unmanned aerial vehicle small target continuous frame coupled time sequence image; Based on the topological equivalent envelope model of the target region and the direction vector equation established with the geometric center of the target interception window, the envelope feature point set of the UAV small target in the first frame and the extraction rules of the envelope feature points in the continuous frames are obtained. Based on the envelope feature point set of the UAV small target in the initial frame and the historical continuous frame UAV small target envelope feature point set extracted according to the extraction rules of continuous frame envelope feature points, the target capture pane of the current frame is obtained; Based on the extracted target envelope in the target capture pane of the current frame prediction and the target envelope of the historical frames, homotopy discrimination and ID consistency evaluation of the small target envelope of the moving UAV in the inter-frame images are performed. Based on the UAV small target detection and prediction results and envelope homotopy criterion, a confidence consistency method is constructed. An equal-interval time-division dual-loop mechanism based on detection and tracking is established to track the motion and behavior of UAV small targets, and realize the ID labeling and stable tracking of inter-frame homotopy targets.

[0011] The present invention also provides a time-division detection and tracking dual-loop device, comprising: The first processing module is used to acquire in real time coupled temporal images of small UAV targets with a combination of long and short focal lengths using ground-to-air image acquisition equipment; The second processing module is used to obtain the topological equivalent envelope model of the target capture pane and the target region based on the first frame image of the coupled temporal image of the continuous frames of the UAV small target. The third processing module is used to obtain the envelope feature point set of the UAV small target in the first frame and the extraction rules of the envelope feature points in the continuous frames based on the topological equivalent envelope model of the target area and the direction vector equation established with the geometric center of the target interception window. The fourth processing module is used to obtain the target capture pane for the current frame prediction based on the envelope feature point set of the UAV small target in the initial frame and the envelope feature point set of the UAV small target in the historical continuous frames extracted according to the extraction rules of the envelope feature points of the continuous frames. The fifth processing module is used to perform homotopy discrimination and ID consistency evaluation of small targets of moving UAVs in inter-frame images based on the extracted target envelope in the target capture pane of the current frame and the target envelope of historical frames. The sixth processing module is used to construct a confidence consistency method based on the detection and prediction results of small targets of UAVs and the envelope homotopy criterion, establish an equal-interval time-division dual-loop mechanism based on detection and tracking, track the motion and behavior of small targets of UAVs, and realize the ID labeling and stable tracking of inter-frame homotopy targets.

[0012] The present invention also provides a time-division detection and tracking dual-loop system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes the time-division detection and tracking dual-loop method when executed by the processor.

[0013] The present invention also provides a storage medium storing a computer program, which executes a time-sharing detection and tracking dual-loop method during runtime.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention achieves high-precision, stable, and real-time tracking of small targets on UAVs in complex scenarios through multi-dimensional technology collaboration. First, it employs topological envelope modeling and point set reconstruction techniques, using the geometric center point of the target capture pane as a reference for isoangular sampling to construct a vector envelope description, replacing the traditional detection box and enhancing the fine-grained representation of small target morphology. Second, it establishes motion observation units based on envelope feature points, modeling the target motion state through vector transfer functions and prediction models, overcoming the sensitivity of traditional detection boxes to appearance deformation and strengthening tracking stability. Third, it introduces a target homotopy discrimination mechanism and a confidence energy equation, utilizing the mapping relationship and intersection-union ratio (IU / UGROUP) calculation of envelope feature point sets between consecutive frames to construct an ID consistency evaluation system, effectively suppressing target re-identification ID jumps caused by occlusion. Fourth, it designs a time-division dual-loop mechanism of "detection ⇋ tracking," where detection frames perform global detection, and tracking frames perform local pane target edge extraction, significantly reducing computational load and collaboratively optimizing accuracy and speed. Finally, the time-sharing detection and tracking dual-loop mechanism supports dynamic adaptation to different detection algorithms and tracking models, breaking through the dependence of traditional frameworks on specific models and improving the system's generalization ability to multi-dimensional environments and cross-modal scenarios. In summary, this invention solves the problems of insufficient small target feature representation, occlusion tracking failure, difficulty in balancing real-time performance and stability, and weak model generalization in existing methods through the synergy of topological representation, motion modeling, homotopy discrimination, and time-sharing dual-loop. Attached Figure Description

[0015] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of the time-sharing detection and tracking dual-loop method according to an embodiment of the present invention; Figure 2 This invention provides a boundary map of a topologically equivalent envelope model of a single-connected UAV small target region containing a set of feature points, generated based on the target capture pane.

[0017] Figure 3 The present invention utilizes a time-division detection and tracking dual-loop method to achieve tracking results in a real video stream. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0020] Example 1 like Figure 1 As shown, the present invention provides a time-division detection and tracking dual-loop method, comprising: S1, continuous frame image acquisition of UAV targets: real-time acquisition of continuous frame coupled temporal images of small UAV targets by combining long and short focal lengths through ground-to-air image acquisition equipment.

[0021] Specifically, radar and other detection equipment are used to perform a full-coverage scan of the low-altitude airspace to search for and locate flying foreign objects. Based on the location results, the flying foreign object is focused into the field of view of a visible light or infrared long- and short-focal-length coupled imaging system, and features of the foreign object within the field of view are extracted and identified. If it is confirmed to be a drone, the camera's line of sight is adjusted by a servo mechanism to stably lock the target at the center of the field of view, and continuous frame images are acquired in real time. The acquired video stream is then transmitted to the back-end processing system. If it is confirmed not to be a drone, the camera remains silent.

[0022] S2, Initial Frame Image Detection and Construction of Topological Space: Based on the initial frame image of the coupled temporal images of the UAV small target in continuous frames, the topological equivalent envelope model of the target capture pane and the target region is obtained.

[0023] Specifically, S2 includes the following steps: Step S201: The target detection algorithm is used to identify the first frame image of the coupled temporal image of the small target UAV in continuous frames, to determine the UAV target and obtain its identity ID and detection confidence. At the same time, the target capture pane is used to identify the target. A To achieve spatial location determination of the target, the target capture pane is as follows: ; in, x and y Indicates the center coordinates of the target capture pane. w and h This indicates the width and height of the target capture pane.

[0024] Step S202: Using the boundary of the target capture pane as a spatial constraint, the image region within the pane is accurately segmented and extracted from the entire image to form an independent target ROI (Region of Interest) processing unit.

[0025] Step S203: In the target ROI processing unit, a topologically equivalent envelope model of the target region is established by topological equivalence of a single frame two-dimensional image.

[0026] ; in, For the current frame's topological equivalent space under temporal conditions, F A single frame image. A Capture the pane area for the target in the image. T This is the set of envelope points for the actual target pixel region in the target segmentation result within the pane. ∈ , It is the set of vectors formed between the geometric center of the target and the sampling points of the target envelope boundary; S3, Envelope Feature Point Set Construction: Based on the topological equivalent envelope model of the target region and the direction vector equation established with the geometric center of the target interception pane, the envelope feature point set of the UAV small target in the initial frame and the extraction rules of the envelope feature points in the continuous frames are obtained.

[0027] Specifically, S3 includes the following steps: Step S301, solving the boundary of the topological equivalent envelope model of the simply connected target region, specifically: First, focus on the target capture pane. A Local images within Solve for the second-order difference gradient value of the subgraph of the region where the target is located. : ; in, express x Directional second-order difference, express y Directional second-order difference; Secondly, for the second-order difference gradient value Binarization is performed to obtain a binary gradient map. : ; Where T is the binarization threshold; Then, by using erosion and dilation as auxiliary methods to adjust the regional connectivity, the adjusted binary map is obtained. : ; Where L represents the structural element of the morphological operation. Indicates corrosion operation. Indicates an expansion operation; Finally, through comparison By calculating the area of ​​each connected region, the region with the largest area is selected as the target region, resulting in a simply connected binary graph. Then, morphological gradients are used to extract the boundary of the target region, resulting in the topological equivalent envelope model boundary of the simply connected target region. ,renew With the target capture pane. Boundary of the topological equivalent envelope model of a simply connected target region. for: ; Step S302, constructing the envelope feature point set of the UAV small target, specifically: First, capture the pane of the target image in the initial frame. A Geometric center coordinates of the region O 1 ( u , v Using 0 as the reference, with the positive direction of the horizontal coordinate axis as position 0, the angle... β To increase the step size, construct a unit vector in a counter-clockwise direction. The point intersects the boundary of the topologically equivalent envelope model of the simply connected target region at point […]. ∈ T ,Right now, Then, iterate through =1 to The envelope feature point set of the small target of the UAV in the first frame is obtained. .

[0028] Step S303, establish the extraction rules for the envelope feature points of consecutive frames, specifically as follows: For subsequent consecutive frames coupled with temporal images of a small UAV target, a target detection algorithm is first used to identify and locate the small UAV target, obtaining its target capture pane. Then, using the pane boundaries as constraints, Regions of Interest (ROIs) are segmented from the entire image. Within each ROI, a topologically equivalent envelope model of the target region is constructed using a two-dimensional image topological equivalence method. Next, by solving the boundary of the simply connected target region topologically equivalent envelope model and combining it with the direction vector equation established with the geometric center of the target capture pane as the origin, the envelope feature point set of the small UAV target is finally obtained. This can be combined with... Figure 2 Understand how to determine the envelope feature points of a single-frame image.

[0029] S4, Target Motion Pattern Analysis and Estimation: Based on the envelope feature point set of the UAV small target in the initial frame and the historical continuous frame UAV small target envelope feature point set extracted according to the extraction rules of continuous frame envelope feature points, the target capture pane of the current frame is obtained.

[0030] Specifically, S4 includes the following steps: Step S401: Extract the envelope feature point set of the UAV small target in subsequent historical consecutive frames according to the extraction rules of continuous frame envelope feature points, and combine it with the envelope feature point set of the UAV small target in the initial frame. The set of corresponding points between subsequent consecutive historical frames is calculated. That is, the set of target envelope feature points in historical continuous frames.

[0031] Step S402, using the set of target envelope feature points from historical consecutive frames. For the observed object, the feature point moves within a continuous image window using a vector. This indicates that, based on the target envelope feature point set... Based on the changing patterns, a prediction system for the envelope feature points of small targets on UAVs was constructed. ; ; in, Represents a vector transfer function. Represents the prediction function. This represents the movement vector of historical frame feature points within a continuous image window, where A and B are the weighting coefficients of the measured and predicted values, respectively, based on the set of historical frame envelope feature points. with point vector Predict all envelope feature points of the small target UAV in the current frame to obtain the envelope feature point set of the current frame. .

[0032] Step S403, determining the target capture pane for the current frame, specifically: First, calculate the envelope feature point set. All feature points within x , y The extreme values ​​of the directions yield the minimum bounding box containing all predicted feature points. : ; Where s represents the total number of feature points in the target envelope. , , , .

[0033] Then, using the minimum bounding box as a reference, expand outwards proportionally. r Expand to get the target capture pane for the current frame. : ; S5, Inter-frame target homotopy discrimination: Based on the extracted target envelope in the target capture pane of the current frame and the target envelope of the historical frames, perform homotopy discrimination and ID consistency evaluation of the small target envelope of the moving UAV in the inter-frame images.

[0034] Specifically, S5 includes the following steps: Step S501: Capture the target pane in the current frame. The internal image is processed using the method in step 301 to obtain the currently frame's simply connected binary image. That is, the envelope of the small drone target extracted in the current frame.

[0035] Step S502, homotopy discrimination of targets between consecutive frames, specifically: Capture the geometric center coordinates of the window frame by target O t ( u , v Taking ) as the base point, because If the target envelope between adjacent frames can be considered as two envelope loops with the same base point, then it can be achieved by: ; Will and Using the original space and image space respectively, continuous full mapping can be achieved, that is: ; in, for arrive A continuous mapping, because For discrete sampling points of the target envelope, therefore, it is always possible to find , so that satisfaction arrive The continuous full screenings.

[0036] Step S503, Inter-frame target ID consistency assessment, specifically: Calculate the intersection-union ratio (IoU) of pixels within the target envelope of the current frame and pixels within the target envelope of historical frames, and construct a confidence energy equation that includes three elements: IoU of pixels within the target envelope of consecutive frames, confidence of the motion law measurement results of each point, and IoU of pixels within the target capture pane of consecutive frames. ; ; ; in, and As weight, The cross-union ratio of pixels within the target envelope of consecutive frames of images. Intersection over Union (IoU) of pixels within the target capture pane in consecutive frame images. n This represents the number of elements within the target envelope point set of a single frame image. The angle step size is... For each point, a confidence score is assigned. The set of pixels within the target envelope. Capture the set of pixels within the target pane, and set a threshold accordingly. By comparing confidence energy values, consistency assessment of small target IDs can be achieved. S6, Time-Division Detection and Tracking Dual Loop: Steps S1 to S5 are the detection and tracking processing steps for the first sequence among multiple equally spaced sequences. Based on the UAV small target detection, prediction results, and envelope homotopy criterion, a confidence consistency method is constructed to establish an equally spaced time-division dual loop mechanism based on detection and tracking. This mechanism tracks the motion and behavior of UAV small targets, achieving ID labeling and stable tracking of inter-frame homotopy targets. It can be combined with... Figure 3 Understand ID calibration and stable tracking based on a detection and tracking-based time-division dual-loop mechanism.

[0037] Specifically, the time-sharing detection and tracking dual-loop mechanism is as follows: A confidence-consistent method based on UAV small target detection, prediction results, and envelope homotopy criterion is established, creating a time-division dual-loop mechanism of "detection ⇋ tracking" at equal intervals, using continuous frame images F as the basis: ; equal-interval keyframes As anchor nodes, with To track nodes, using the same time interval as a benchmark, the target detection algorithm only... Detect keyframes and The confidence scores of the target detection results are compared. If the confidence score increases, the target is assigned a new label and ID based on the new detection result; otherwise, the label and ID remain unchanged. For tracking nodes, only inter-frame tracking is performed, thereby completing the time-division dual-loop mechanism for UAV target detection and tracking.

[0038] This invention provides a target tracking technology that achieves real-time, continuous, and stable tracking of small UAV targets with occlusion in complex scenarios through the synergistic application of multi-dimensional technologies, including topological representation, motion modeling, homotopy discrimination, and a dual-loop mechanism. First, an initial frame detection is performed on the acquired real-time video stream to construct a topologically equivalent envelope model of the target capture pane and the target region, laying the foundation for subsequent operations. Within the topologically equivalent envelope model of the target region, a vector equation is established starting from the geometric center point of the target capture pane to determine the envelope feature points. Based on the inter-frame temporal relationship between the target envelope feature points and the envelope center point, an envelope sampling point prediction system is derived to predict the position of the target envelope feature points in the current frame, thereby obtaining the predicted target capture pane for the current frame. The extracted target envelope in the predicted target capture pane of the current frame is compared with the target envelopes of historical frames to achieve homotopy discrimination and ID continuation; simultaneously, the feature points of the extracted envelope are used to calibrate the envelope feature point prediction system. Subsequent frame images repeat the above steps to eliminate the influence of occlusion and achieve real-time, continuous tracking of the UAV. In this process, video frames are re-detected at equal intervals during different time periods, and the detection results are assigned to the tracking process to realize the dynamic updating of the envelope feature point prediction system and the identity ID. Thus, the detection and tracking processes constitute a "detection ⇋ tracking" dual-loop mechanism. Through dynamic iterative verification and closed-loop feedback between the two throughout the entire tracking cycle, the target tracking accuracy and system robustness are significantly improved.

[0039] Example 2 The present invention also provides a time-division detection and tracking dual-loop device, comprising: The first processing module is used to acquire in real time coupled temporal images of small UAV targets with a combination of long and short focal lengths using ground-to-air image acquisition equipment; The second processing module is used to obtain the topological equivalent envelope model of the target capture pane and the target region based on the first frame image of the coupled temporal image of the continuous frames of the UAV small target. The third processing module is used to obtain the envelope feature point set of the UAV small target in the first frame and the extraction rules of the envelope feature points in the continuous frames based on the topological equivalent envelope model of the target area and the direction vector equation established with the geometric center of the target interception window. The fourth processing module is used to obtain the target capture pane for the current frame prediction based on the envelope feature point set of the UAV small target in the initial frame and the envelope feature point set of the UAV small target in the historical continuous frames extracted according to the extraction rules of the envelope feature points of the continuous frames. The fifth processing module is used to perform homotopy discrimination and ID consistency evaluation of small targets of moving UAVs in inter-frame images based on the extracted target envelope in the target capture pane of the current frame and the target envelope of historical frames. The sixth processing module is used to construct a confidence consistency method based on the detection and prediction results of small targets of UAVs and the envelope homotopy criterion, establish an equal-interval time-division dual-loop mechanism based on detection and tracking, track the motion and behavior of small targets of UAVs, and realize the ID labeling and stable tracking of inter-frame homotopy targets.

[0040] Example 3 The present invention also provides a time-division detection and tracking dual-loop system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes the time-division detection and tracking dual-loop method when executed by the processor.

[0041] Example 4 The present invention also provides a storage medium storing a computer program, which executes a time-sharing detection and tracking dual-loop method during runtime.

[0042] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A time-division detection and tracking dual-loop method, characterized in that, include: Real-time acquisition of continuous frame coupled temporal images of small targets on UAVs by combining long and short focal lengths using ground-to-air image acquisition equipment; Based on the first frame image of the coupled temporal images of the UAV small target in consecutive frames, a topologically equivalent envelope model of the target capture pane and the target region is obtained; Based on the topological equivalent envelope model of the target region and the direction vector equation established with the geometric center of the target interception window, the envelope feature point set of the UAV small target in the first frame and the extraction rules of the envelope feature points in the continuous frames are obtained. Based on the envelope feature point set of the UAV small target in the initial frame and the historical continuous frame UAV small target envelope feature point set extracted according to the extraction rules of continuous frame envelope feature points, the target capture pane of the current frame is obtained; Based on the extracted target envelope in the target capture pane of the current frame prediction and the target envelope of the historical frames, homotopy discrimination and ID consistency evaluation of the small target envelope of the moving UAV in the inter-frame images are performed. Based on the UAV small target detection and prediction results and envelope homotopy criterion, a confidence consistency method is constructed. An equal-interval time-division dual-loop mechanism based on detection and tracking is established to track the motion and behavior of UAV small targets, and realize the ID labeling and stable tracking of inter-frame homotopy targets.

2. A time-division detection and tracking dual-loop device, characterized in that, include: The first processing module is used to acquire in real time coupled temporal images of small UAV targets with a combination of long and short focal lengths using ground-to-air image acquisition equipment; The second processing module is used to obtain the topological equivalent envelope model of the target capture pane and the target region based on the first frame image of the coupled temporal image of the continuous frames of the UAV small target. The third processing module is used to obtain the envelope feature point set of the UAV small target in the first frame and the extraction rules of the envelope feature points in the continuous frames based on the topological equivalent envelope model of the target area and the direction vector equation established with the geometric center of the target interception window. The fourth processing module is used to obtain the target capture pane for the current frame prediction based on the envelope feature point set of the UAV small target in the initial frame and the envelope feature point set of the UAV small target in the historical continuous frames extracted according to the extraction rules of the envelope feature points of the continuous frames. The fifth processing module is used to perform homotopy discrimination and ID consistency evaluation of small targets of moving UAVs in inter-frame images based on the extracted target envelope in the target capture pane of the current frame and the target envelope of historical frames. The sixth processing module is used to construct a confidence consistency method based on the detection and prediction results of small targets of UAVs and the envelope homotopy criterion, establish an equal-interval time-division dual-loop mechanism based on detection and tracking, track the motion and behavior of small targets of UAVs, and realize the ID labeling and stable tracking of inter-frame homotopy targets.

3. A time-division detection and tracking dual-loop system, characterized in that, include: A memory and a processor, wherein the memory stores a computer program executed by the processor, the computer program performing the time-sharing detection and tracking dual-loop method as described in claim 1 when executed by the processor.

4. A storage medium, characterized in that, The storage medium stores a computer program, which executes the time-sharing detection and tracking dual-loop method as described in claim 1 when it runs.

Images