A plimpton wild horse recognition system and method based on video tracking, processing device and storage medium

Abstract

The present application relates to a kind of pronghorn identification system, method, processing device and storage medium based on video tracking, the system includes: data input interface, for receiving the video or image data stream from one or more camera equipment;Data type judging module is used to judge the data type of video or image data stream based on the meta information of video or image data stream;Refinement individual analysis channel is used to accurately track the individual of pronghorn in video or image data stream, trajectory optimization and behavior analysis are carried out to near view data, and the refinement individual analysis result of pronghorn is obtained;Macroscopic group analysis channel is used to carry out population statistics and rhythm analysis to long-range or aerial data in video or image data stream, and the macroscopic group analysis result of pronghorn is obtained;Data output interface is used to output the refinement individual analysis result and macroscopic group analysis result of pronghorn, the present application can be widely applied in computer vision and intelligent monitoring technical field.

Description

Technical Field

[0001] This invention relates to the fields of computer vision and intelligent monitoring technology, and in particular to a Przewalski's horse recognition system, method, processing device, and storage medium based on video tracking. Background Technology

[0002] Protected areas are the cornerstone of global conservation of endangered species, and their effective management relies on timely and accurate monitoring data. Traditional wildlife monitoring methods mainly include: 1) Physical tagging, such as fitting animals with GPS collars. While these methods can provide individual location information, they have many drawbacks: First, the capture and tagging process is highly invasive and may cause physiological and behavioral disturbances to animals; second, the equipment is expensive and has limited battery life, making it difficult to popularize and conduct long-term monitoring in large populations. 2) Manual survey methods, such as transect methods. These methods rely on manual field observation, are time-consuming and labor-intensive, and are difficult to achieve high-frequency coverage in vast, topographically complex protected areas. 3) Biological sampling methods, such as fecal DNA analysis. While these methods can provide genetic and population information, they also suffer from high costs, logistical challenges, and a severe lack of real-time data. In summary, these traditional methods have significant bottlenecks in monitoring costs, animal welfare, and data spatiotemporal resolution, and cannot meet the refined and real-time requirements of modern scientific management of protected areas.

[0003] To overcome the limitations of traditional methods, deep learning-based computer vision technology has gained widespread application as a non-invasive monitoring tool. For example, existing technology discloses a multifunctional mammalian ecological monitoring system, method, processing equipment, and storage medium based on drones. This method uses object detection networks to automatically identify and count the number of animals in aerial images, improving census efficiency. However, such methods are only suitable for macroscopic population statistics and cannot analyze the fine-grained behaviors of Przewalski's horses (such as socialization and drinking).

[0004] In summary, existing computer vision-based monitoring methods are often "single-paradigm," focusing either on macroscopic population surveys or microscopic individual identification, lacking a unified framework capable of handling both tasks simultaneously. However, in real protected areas, monitoring data is inherently multi-scale, including both wide-angle views of water sources (for statistical analysis of population rhythms) and close-up shots (for analyzing individual behavior). Existing "single-scale" models cannot efficiently and reliably handle this mixed data stream. Furthermore, during individual tracking, existing visual tracking algorithms (such as DeepSORT or BoT-SORT) are prone to trajectory interruptions and identity (ID) jumps in complex field environments, particularly when Przewalski's horses frequently cross paths, experience prolonged occlusion, or rapidly enter or exit the frame. This leads to severely inaccurate subsequent behavioral analysis data (such as behavioral time estimation). Summary of the Invention

[0005] To address the aforementioned problems, the purpose of this invention is to provide a Przewalski's horse identification system, method, processing device, and storage medium based on video tracking, which can significantly improve the stability and accuracy of individual tracking in complex field environments.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: Firstly, it provides a Przewalski's horse recognition system based on video tracking, comprising: A data input interface for receiving video or image data streams from one or more camera devices; The data type determination module is used to determine the data type of a video or image data stream based on its metadata. The refined individual analysis channel is used to accurately track, optimize trajectories, and analyze behavior of Przewalski's horses in close-up data from video or image data streams, resulting in refined individual analysis results of Przewalski's horses. The macro-population analysis channel is used to perform population statistics and rhythm analysis on distant or aerial data in video or image data streams to obtain macro-population analysis results of Przewalski's horses; The data output interface is used to output refined individual analysis results and macro-group analysis results of Przewalski's horses to solve the problems of individual tracking under complex occlusion and group dynamic analysis in large-scale scenes.

[0007] Furthermore, the refined individual analysis channel includes: The target detection module is used to use a pre-trained target detection model to detect all Przewalski's horse individuals in the video frames of the video or image data stream sent by the data input interface, and output the bounding box coordinates of each Przewalski's horse individual target. The multi-target tracking module is used to predict the motion state and associate data of individual Przewalski's horses based on the bounding box coordinates of the individual Przewalski's horses using a multi-target tracking algorithm. It assigns a temporary identity to each individual Przewalski's horse that appears in a video frame and outputs the original tracking trajectory of each individual Przewalski's horse. The trajectory optimization module is used to determine whether each Przewalski's horse individual target has experienced an occlusion loss event based on the original tracking trajectory of each Przewalski's horse individual target, and to obtain the optimized tracking trajectory of each Przewalski's horse individual target by using an occlusion-location association matching method. The individual behavior analysis module is used to classify the behavior of each individual Przewalski's horse based on the optimized tracking trajectory and corresponding video clips, and on the kinematic and / or spatiotemporal video features of the individual Przewalski's horse. The module outputs the duration and time interval of each individual Przewalski's horse performing a certain behavior.

[0008] Furthermore, the trajectory optimization module includes: The trajectory and occlusion monitoring unit is used to monitor the trajectory status of all Przewalski's horse individual targets in each video frame in real time and calculate the degree of overlap between the bounding boxes of each Przewalski's horse individual target. The occlusion loss event determination unit is used to determine whether an occlusion loss event has occurred for each Przewalski's horse individual target based on the degree of overlap between the bounding boxes of each Przewalski's horse individual target and a pre-set overlap threshold. The occlusion context recording unit is used to mark the identity of the Przewalski's horse individual target that has experienced an occlusion loss event as a waiting reconnection state, and at the same time record the identity of the Przewalski's horse individual target associated with the Przewalski's horse individual target that has experienced an occlusion loss event and its current original tracking trajectory, and use the associated Przewalski's horse individual target as the associated occlusion object of the Przewalski's horse individual target that has experienced an occlusion loss event. The new trajectory discovery unit is used to assign a new identity to a new Przewalski's horse individual target when a new Przewalski's horse individual target appears that cannot be matched with any original tracking trajectory; The occlusion-location association matching unit is used to calculate the initial position of the new Przewalski's horse individual target based on the data recorded by the occlusion context recording unit when the new Przewalski's horse individual target appears, and to calculate the spatial distance between the target and the current position of the occluding object individual, wherein the occluding object individual is the Przewalski's horse individual target associated with the Przewalski's horse individual target that has experienced an occlusion loss event; The identity reconnection unit is used to determine whether the new Przewalski's horse individual target is an occluded individual based on the calculated spatial distance and a preset spatial distance threshold. If so, the identity reconnection operation is performed to correct the identity of the new Przewalski's horse individual target to that of the occluded individual, clear the waiting reconnection state of the occluded individual, and merge the two tracking trajectories into one optimized tracking trajectory. The occluded individual is the Przewalski's horse individual target that has experienced an occlusion loss event.

[0009] Furthermore, the macro-group analysis channel includes: The population statistics module is used to automatically detect and count the total number of Przewalski's horses in distant views or aerial data in video or image data streams using target detection models or density map estimation algorithms, and output "time-quantity" data pairs. The group rhythm analysis module is used to aggregate and analyze the rhythm of time-quantity data pairs, and output the activity rhythm diagram of the animal group and related statistical reports.

[0010] Secondly, a Przewalski's horse identification method based on video tracking is provided, including: Receive video or image data streams from one or more camera devices; Determine the data type of the video or image data stream based on its metadata. When the data type of the video or image data stream is close-up data, the Przewalski's horse individual is accurately tracked, its trajectory optimized, and its behavior analyzed to obtain refined individual analysis results of the Przewalski's horse. When the data type of the video or image data stream is a distant view or aerial photography data, population statistics and rhythm analysis are performed on the video or image data stream to obtain the macroscopic population analysis results of Przewalski's horses; It outputs refined individual analysis results and macro-group analysis results of Przewalski's horses to solve the problems of individual tracking under complex occlusion and group dynamic analysis in large-scale scenes.

[0011] Furthermore, when the data type of the video or image data stream is close-up data, the video or image data stream is subjected to precise tracking, trajectory optimization, and behavioral analysis of individual Przewalski's horses to obtain refined individual analysis results of Przewalski's horses, including: A pre-trained target detection model is used to detect all Przewalski's horse individuals in video frames of video or image data streams, and output the bounding box coordinates of each Przewalski's horse individual target. A multi-target tracking algorithm is adopted. Based on the bounding box coordinates of individual Przewalski's horse targets, the motion state of individual Przewalski's horse targets is predicted and data is associated. A temporary identity is assigned to each individual Przewalski's horse target that appears in the video frame, and the original tracking trajectory of each individual Przewalski's horse target is output. Based on the original tracking trajectory of each Przewalski's horse individual target, it is determined whether each Przewalski's horse individual target has experienced an occlusion loss event, and the occlusion-location association matching method is used to obtain the optimized tracking trajectory of each Przewalski's horse individual target. Based on the optimized tracking trajectory of each Przewalski's horse individual target and the corresponding video or image data stream, the behavior of each individual target is classified according to the kinematic characteristics and / or spatiotemporal video characteristics, and the duration and time interval of each individual Przewalski's horse performing a certain behavior are output.

[0012] Furthermore, based on the original tracking trajectory of each Przewalski's horse individual target, determining whether an occlusion loss event has occurred for each Przewalski's horse individual target, and using an occlusion-location association matching method to obtain the optimized tracking trajectory of each Przewalski's horse individual target, includes: The trajectory status of all Przewalski's horse individual targets in each video frame is monitored in real time, and the degree of overlap between the bounding boxes of each Przewalski's horse individual target is calculated. Based on the degree of overlap between the bounding boxes of each Przewalski's horse individual target and a pre-set overlap threshold, it is determined whether each Przewalski's horse individual target has experienced an occlusion loss event. When an occlusion loss event is triggered, the identity of the Przewalski's horse individual target that experienced the occlusion loss event is marked as waiting for reconnection and added to the waiting reconnection list. At the same time, the identity of the Przewalski's horse individual target associated with the Przewalski's horse individual target that experienced the occlusion loss event and its current original tracking trajectory are recorded, and the associated Przewalski's horse individual target is used as the associated occlusion object of the Przewalski's horse individual target that experienced the occlusion loss event. Monitor whether any new Przewalski's horse individual target tracks are initialized. When a new Przewalski's horse individual target exists, assign a new identity to the new Przewalski's horse individual target and trigger occlusion-location association matching. Based on the waiting reconnection list and the identity of the Przewalski's horse individual targets associated with the occlusion loss event and their current original tracking trajectories, occlusion-location association matching is performed. Based on the calculated spatial distance and the preset spatial distance threshold, it is determined whether the new individual is an occluded individual. If so, an identity reconnection operation is performed to correct the identity of the new individual to that of the occluded individual, clear its waiting reconnection state, and merge the two tracking trajectories into one optimized tracking trajectory.

[0013] Furthermore, when the data type of the video or image data stream is a distant view or aerial photography data, population statistics and rhythm analysis are performed on the video or image data stream to obtain macroscopic population analysis results of Przewalski's horses, including: Using target detection models or density map estimation algorithms, the total number of Przewalski's horses in distant views or aerial data in video or image data streams is automatically detected and counted, and "time-count" data pairs are output. Data aggregation and rhythm analysis are performed on the "time-quantity" data to output the activity rhythm diagram of the animal group and related statistical reports.

[0014] Thirdly, a processing device is provided, including computer program instructions, wherein when the computer program instructions are executed by the processing device, they are used to implement the steps corresponding to the above-described video tracking-based Przewalski's horse recognition method.

[0015] Fourthly, a computer-readable storage medium is provided, wherein computer program instructions are stored on the computer-readable storage medium, wherein the computer program instructions, when executed by a processor, are used to implement the steps corresponding to the above-described video tracking-based Przewalski's horse recognition method.

[0016] The present invention has the following advantages due to the adoption of the above technical solutions: 1. This invention constructs a unified analysis framework by creating parallel refined individual analysis channels and macro group analysis channels, which solves the problem of multi-scale data incompatibility and enables comprehensive and multi-level analysis of monitoring scenarios.

[0017] 2. The analytical framework constructed in this invention can simultaneously extract two different scales of information with high ecological relevance from multi-source heterogeneous data collected from the same monitoring area (such as a water source): one is the precise behavior at the individual level, and the other is the macroscopic dynamics at the group level. This design overcomes the shortcomings of existing technologies that cannot take into account both macroscopic and microscopic aspects, enabling researchers to not only understand the number of individuals in distant groups (group statistics), but also to deeply analyze the specific behaviors of animals under close monitoring (individual behavior), thus providing an unprecedented comprehensive insight into the social ecology of species.

[0018] 3. Existing tracking algorithms are prone to trajectory interruptions and identity changes when dealing with frequent crossings and occlusions between animals. This invention incorporates a trajectory optimization module, the core of which is occlusion-location association matching logic. This significantly improves the robustness and accuracy of multi-target tracking in complex occlusion environments, ensuring the reliability of downstream behavior analysis.

[0019] 4. The occlusion-location association matching logic set up in this invention no longer relies solely on unstable motion model predictions (such as Kalman filtering), but cleverly utilizes the identity and location of the occluded individual as stable spatiotemporal context anchors. When an occluded individual reappears after being lost due to occlusion, this invention can accurately reconnect its identity to its original trajectory based on its spatial proximity to the occluded individual. This reconnection mechanism based on occlusion context greatly reduces identity jump errors caused by high-frequency occlusion in complex social and densely gathered scenarios in the wild (such as drinking water at a water source).

[0020] In summary, this invention can be widely applied in the fields of computer vision and intelligent monitoring technology. Attached Figure Description

[0021] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. In the drawings: Figure 1 This is a schematic diagram of a multi-scale visual analysis framework provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the location reconnection framework provided in an embodiment of the present invention. Detailed Implementation

[0022] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art.

[0023] It should be understood that the terminology used herein is for the purpose of describing particular exemplary embodiments only and is not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “described” as used herein may also include the plural forms. The terms “comprising,” “including,” “containing,” and “having” are inclusive and therefore indicate the presence of the stated features, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, elements, components, and / or combinations thereof. The method steps, processes, and operations described herein are not construed as requiring them to be performed in a particular order described or illustrated unless the order of performance is explicitly indicated. It should also be understood that additional or alternative steps may be used.

[0024] Although terms such as first, second, third, etc., may be used in this document to describe multiple elements, components, regions, layers, and / or segments, these elements, components, regions, layers, and / or segments should not be limited by these terms. These terms may be used only to distinguish one element, component, region, layer, or segment from another. Unless the context clearly indicates otherwise, terms such as "first," "second," and other numerical terms used herein do not imply order or sequence. Therefore, the first element, component, region, layer, or segment discussed below may be referred to as the second element, component, region, layer, or segment without departing from the teachings of the exemplary embodiments.

[0025] Existing technologies either focus on macro-level population surveys or micro-level individual identification, lacking a unified framework capable of processing multi-scale video data streams in parallel, ranging from wide-angle views to close-up details. This results in low data utilization and an inability to simultaneously meet the dual analytical needs of both groups and individuals. Furthermore, existing visual tracking algorithms suffer from poor robustness and tracking failures in complex field environments. In real-world scenarios, Przewalski's horses frequently occlude each other, cross paths, and rapidly enter and exit the frame, causing traditional tracking algorithms to easily experience trajectory interruptions and identity ID jumps, severely impacting the accuracy of subsequent individual behavior analysis. This invention provides a Przewalski's horse identification system based on video tracking, comprising: a data input interface for receiving video or image data streams from one or more camera devices; a data type determination module for determining the data type of the video or image data stream based on its metadata; a refined individual analysis channel for performing precise tracking, trajectory optimization, and behavioral analysis of Przewalski's horse individuals on close-up data in the video or image data stream, obtaining refined individual analysis results; a macro-group analysis channel for performing population statistics and rhythm analysis on distant or aerial data in the video or image data stream, obtaining macro-group analysis results; and a data output interface for outputting the refined individual analysis results and macro-group analysis results of Przewalski's horse, thereby solving the problems of individual tracking under complex occlusion and group dynamic analysis in large-scale scenes. This invention aims to efficiently process multi-source video data through parallel analysis of refined individuals and macro-groups, and, combined with a novel location reconnection algorithm, significantly improve the stability and accuracy of individual tracking in complex wild environments, thus providing reliable technical support for the study of individual behavior and group rhythms of wild animals.

[0026] Example 1 like Figure 1 As shown, this embodiment provides a Przewalski's horse recognition system based on video tracking, including a data input interface, a data type judgment module, a refined individual analysis channel, a macro-group analysis channel, and a data output interface.

[0027] The data input interface is used to receive video or image data streams from one or more camera devices.

[0028] The data type determination module is used to determine the data type of a video or image data stream based on its metadata.

[0029] The refined individual analysis channel is used to accurately track, optimize trajectories, and analyze behavior of Przewalski's horses in close-up data from video or image data streams (such as video or image data streams taken by close-up cameras at water sources). The refined individual analysis results of Przewalski's horses include the optimized tracking trajectory of each individual Przewalski's horse and the duration and time interval of each individual Przewalski's horse performing a certain behavior.

[0030] The macro-population analysis channel is used to perform population statistics and rhythm analysis on distant or aerial data in video or image data streams (such as video or image data streams from high-altitude panoramic cameras or drone aerial photography) to obtain macro-population analysis results of Przewalski's horses.

[0031] The data output interface is used to output refined individual analysis results and macro-group analysis results of Przewalski's horses to solve the problems of individual tracking under complex occlusion and group dynamic analysis in large-scale scenes.

[0032] In a preferred embodiment, the camera device may be a close-up infrared camera, a long-range wide-angle camera, or a drone, etc.

[0033] In a preferred embodiment, such as Figure 2 As shown, the refined individual analysis channel includes a target detection module, a multi-target tracking module, a trajectory optimization module, and an individual behavior analysis module.

[0034] The target detection module uses a pre-trained target detection model (e.g., YOLOv8 model) to detect all Przewalski's horse individuals in the video frames of the video or image data stream sent by the data input interface, and outputs the bounding box coordinates of each Przewalski's horse individual target.

[0035] The multi-target tracking module is used to employ multi-target tracking algorithms (such as the BoT-SORT algorithm) to predict the motion state and associate data of individual Przewalski's horses based on the bounding box coordinates of the individual Przewalski's horses. It assigns a temporary identity (ID) to each individual Przewalski's horse that appears in a video frame and outputs the original tracking trajectory of each individual Przewalski's horse.

[0036] The trajectory optimization module is used to determine whether each Przewalski's horse individual target has experienced an occlusion loss event based on the original tracking trajectory of each individual Przewalski's horse target, and to obtain the optimized tracking trajectory of each individual Przewalski's horse target by using an occlusion-location association matching method.

[0037] The individual behavior analysis module is used to classify the behavior of each Przewalski's horse individual based on the optimized tracking trajectory of each individual target and the corresponding video segments (i.e., the image sequence of the bounding box of the identity in consecutive video frames), based on the kinematic characteristics (such as speed and orientation) and / or spatiotemporal video features (such as using a 3D-CNN network) of the individual target. The module outputs the duration and time interval of each individual target's behavior.

[0038] In a preferred embodiment, the trajectory optimization module can solve the problems of trajectory interruption and ID jump caused by occlusion or rapid movement. The trajectory optimization module includes a trajectory and occlusion monitoring unit, an occlusion loss event judgment unit, an occlusion context recording unit, a new trajectory occurrence judgment unit, an occlusion-location association matching unit, and an identity reconnection unit.

[0039] The trajectory and occlusion monitoring unit is used to monitor the position of all Przewalski's horse individual targets in each video frame in real time. Multiple video frames form the trajectory status of all Przewalski's horse individual targets, and calculate the degree of overlap between the bounding boxes of each Przewalski's horse individual target.

[0040] The occlusion loss event determination unit is used to determine whether an occlusion loss event has occurred for each Przewalski's horse individual target based on the degree of overlap between the bounding boxes of each Przewalski's horse individual target and a pre-set overlap threshold.

[0041] The occlusion context recording unit is used to mark the identity of the Przewalski's horse individual target (i.e. the occluded individual) that has experienced an occlusion loss event as a waiting reconnection state. At the same time, it records the identity of the Przewalski's horse individual target (i.e. the occluder individual) associated with the Przewalski's horse individual target that has experienced an occlusion loss event and its current original tracking trajectory, and regards the associated Przewalski's horse individual target as the associated occluder of the Przewalski's horse individual target that has experienced an occlusion loss event.

[0042] The new trajectory occurrence judgment unit is used to assign a new identity to a new Przewalski's horse individual target (i.e., a new individual) that cannot be matched with any original tracking trajectory, and to trigger the matching logic occlusion-location association matching unit.

[0043] The occlusion-location association matching unit is used to calculate the initial position of a new individual based on the data recorded by the occlusion context recording unit, and to calculate the spatial distance between the new individual and the current position of the occluding individual.

[0044] The identity reconnection unit is used to determine whether the new individual is an occluded individual based on the calculated spatial distance and the preset spatial distance threshold. If so, the identity reconnection operation is performed to correct the identity of the new individual to that of the occluded individual, clear the waiting reconnection state of the occluded individual, and merge the two tracking trajectories into a single, continuous optimized tracking trajectory.

[0045] Specifically, determining whether an occlusion loss event has occurred includes: when the overlap between the bounding box of one Przewalski's horse individual target (i.e., the occluded individual) and the bounding box of another Przewalski's horse individual target (i.e., the occluded individual) is greater than a preset overlap threshold, it is determined that a high overlap has occurred, and when the tracking trajectory of the occluded individual is interrupted (i.e. lost) in a subsequent video frame, it is determined that an occlusion loss event has occurred for the occluded individual.

[0046] In a preferred embodiment, the macro-group analysis channel includes a group size statistics module and a group rhythm analysis module.

[0047] The population statistics module is used to automatically detect and count the total number of Przewalski's horses in distant or aerial data in video or image data streams using object detection models (such as YOLOv8) or density map estimation algorithms, and outputs "time-count" data pairs.

[0048] The group rhythm analysis module is used to aggregate and analyze the rhythm of time-quantity data pairs, and output the activity rhythm diagram of the animal group and related statistical reports (such as peak periods and rhythm intensity).

[0049] Specifically, data aggregation involves aggregating "time-quantity" data pairs according to a preset time unit (e.g., hourly) to generate time-series data for analysis. Rhythm analysis employs time-series analysis and / or cyclic statistics methods (e.g., using the circular package in R language for diurnal activity rhythm analysis) to analyze the time-series data, quantify the activity patterns of animal groups at specific locations (e.g., water sources), and ultimately output the animal group's activity rhythm diagram and related statistical reports (e.g., peak periods, rhythm intensity).

[0050] Example 2 This embodiment provides a Przewalski's horse identification method based on video tracking, including the following steps: 1) Receive video or image data streams from one or more camera devices.

[0051] 2) Determine the data type of the video or image data stream based on its metadata (e.g., camera number, focal length, installation location).

[0052] 3) When the data type of the video or image stream is close-up data, perform precise tracking, trajectory optimization, and behavioral analysis on individual Przewalski's horses in the video or image stream to obtain refined individual analysis results of Przewalski's horses, specifically: 3.1) A pre-trained target detection model is used to detect all Przewalski's horse individuals in the video frames of the video or image data stream, and output the bounding box coordinates of each Przewalski's horse individual target: 3.1.1) A pre-trained YOLOv8-m object detection model is preloaded. This model has been pre-trained on a large public dataset (such as COCO) and fine-tuned using a private dataset containing tens of thousands of Przewalski's horse bounding boxes to make it have high recall and high precision for detecting wild horses.

[0053] 3.1.2) Input a frame of video or image data stream into the pre-trained YOLOv8-m object detection model. The model performs forward propagation and outputs the bounding box coordinates of all detected Przewalski's horse individuals in the video frame, obtaining a list of bounding box coordinates (e.g., a set of [x, y, width, height]).

[0054] 3.2) A multi-target tracking algorithm is adopted. Based on the bounding box coordinates of individual Przewalski's horse targets, motion state prediction and data association are performed on individual Przewalski's horse targets. A temporary identity is assigned to each individual Przewalski's horse target that appears in the video frame, and the original tracking trajectory of each individual Przewalski's horse target is output.

[0055] Specifically, the multi-target tracking algorithm in this embodiment adopts the BoT-SORT tracking algorithm. This algorithm maintains a list of bounding box coordinates output by the target detection module. Each trajectory includes a unique identity and a motion state based on a Kalman filter. The specific process of this step is as follows: 3.2.1) Using a Kalman filter, the position of each trajectory in the previous frame of the survival trajectory list of the BoT-SORT tracking algorithm is predicted to obtain its predicted position in the current frame, and thus the predicted trajectory is obtained.

[0056] 3.2.2) Calculate the intersection-union (IoU) cost matrix between all predicted trajectories and the new detection boxes in the current frame (i.e., the bounding box coordinates of the Przewalski's horse individual targets detected in the current frame).

[0057] 3.2.3) Using the Hungarian algorithm or a greedy algorithm, all predicted trajectories are matched with the new detection boxes in the current frame based on the calculated intersection-union cost matrix. If the match is successful, the state is updated by using a Kalman filter combined with the position of the new detection box. If the match is unsuccessful, the new detection box is initialized as a new trajectory, i.e., a new identity is assigned. Surviving trajectories that fail to match for multiple consecutive frames are marked as lost.

[0058] 3.2.4) Output all the matched trajectories as the original tracking trajectory for each Przewalski's horse individual target. The original tracking trajectory includes the identity of the corresponding Przewalski's horse individual target. The original tracking trajectory data is discontinuous or prone to jumps when occlusion occurs.

[0059] 3.3) Based on the original tracking trajectory of each Przewalski's horse individual target, determine whether an occlusion loss event has occurred for each individual target, and use an occlusion-location association matching method to obtain the optimized tracking trajectory of each individual Przewalski's horse target: 3.3.1) Monitor the trajectory status of all Przewalski's horse individual targets in each video frame in real time, and calculate the degree of overlap between the bounding boxes of each Przewalski's horse individual target.

[0060] Specifically, all original tracking trajectories in each video frame are monitored in real time, and the intersection-over-union ratio (IoU) between the bounding box of each Przewalski's horse individual target and the bounding boxes of other Przewalski's horse individual targets is calculated.

[0061] 3.3.2) Based on the degree of overlap between the bounding boxes of each Przewalski's horse individual target and a pre-set overlap threshold, determine whether each Przewalski's horse individual target has experienced an occlusion loss event.

[0062] Specifically, when the intersection-over-union (IoU) of the bounding box of the Przewalski's horse individual target (occluded individual) corresponding to a certain original tracking trajectory and the bounding box of the Przewalski's horse individual target (occluded individual) corresponding to another original tracking trajectory is greater than a preset overlap threshold (e.g., 0.6), and the tracking trajectory of the occluded individual is interrupted in a subsequent video frame, it is determined that an occlusion loss event has occurred for the occluded individual.

[0063] 3.3.3) When an occlusion loss event is triggered, the identity of the Przewalski's horse individual target that experienced the occlusion loss event is marked as waiting for reconnection and added to the waiting reconnection list. At the same time, the identity of the Przewalski's horse individual target associated with the Przewalski's horse individual target that experienced the occlusion loss event and its current original tracking trajectory are recorded. The associated Przewalski's horse individual target is used as the associated occlusion object of the Przewalski's horse individual target that experienced the occlusion loss event.

[0064] 3.3.4) Monitor whether any new Przewalski's horse individual target tracks are initialized. When a new Przewalski's horse individual target exists, assign a new identity to the new Przewalski's horse individual target and trigger occlusion-location association matching.

[0065] 3.3.5) Based on the waiting reconnection list and the identity of the Przewalski's horse individual targets associated with the Przewalski's horse individual targets that have experienced occlusion loss events, as well as their current original tracking trajectories, perform occlusion-location association matching.

[0066] 3.3.6) Based on the calculated spatial distance and the preset spatial distance threshold, determine whether the new individual is an occluded individual. If so, perform an identity reconnection operation to correct the identity of the new individual to that of the occluded individual, clear its waiting reconnection state, and merge the two tracking trajectories into a single, continuous optimized tracking trajectory.

[0067] Specifically, when a new Przewalski's horse individual target is created, the following matching logic is executed: A) Traverse the list of objects waiting to reconnect, check if there is an identity waiting to reconnect, and if so, obtain the identity of its associated occluder.

[0068] B) Obtain the initial position of the new Przewalski's horse individual target and the current position of the occluder individual, and calculate the spatial distance between the two positions (e.g., the Euclidean distance between the center points of the two bounding boxes).

[0069] C) If the calculated spatial distance is less than a preset spatial distance threshold (e.g., 200 pixels), then the new Przewalski's horse individual target is determined to be a continuation of the occluded individual, and an identity reconnection operation is performed to correct the identity of the new Przewalski's horse individual target to the identity of the occluded individual.

[0070] D) Remove the occluded individual from the waiting reconnection list and output a set of identity-continuous, robust, optimized tracking trajectories.

[0071] 3.4) Based on the optimized tracking trajectory of each Przewalski's horse individual target and the corresponding video or image data stream, and based on the kinematic characteristics and / or spatiotemporal video characteristics of the individual Przewalski's horse targets, classify the behavior of each identity, and output the duration and time interval of each Przewalski's horse individual performing a certain behavior: 3.4.1) Load a pre-trained video behavior classification model (e.g., I3D or SlowFast 3D-CNN model) that has been pre-trained on a large action recognition dataset (e.g., Kinetics) and fine-tuned with a pre-labeled wild horse behavior dataset (including categories such as drinking, eating, standing, running, and alertness).

[0072] 3.4.2) For each optimized tracking trajectory, extract its corresponding video segment (i.e., the image sequence of the bounding box of the identity in consecutive video frames) from the original video or image data stream.

[0073] 3.4.3) Input the video clip into the pre-trained video behavior classification model for classification, and record the behavior category of the Przewalski's horse individual target corresponding to the optimized tracking trajectory in each video frame.

[0074] 3.4.4) Statistically analyze all video frame records to calculate the duration and time interval of a certain behavior for each individual Przewalski's horse.

[0075] 4) When the data type of the video or image stream is a distant view or aerial photography, perform population statistics and rhythm analysis on the video or image data stream to obtain the macroscopic population analysis results of Przewalski's horses, specifically: 4.1) Using an object detection model or density map estimation algorithm, automatically detect and count the total number of Przewalski's horses in distant views or aerial data in video or image data streams, and output "time-count" data pairs: 4.1.1) In view of the situation where animals in the water source area are highly concentrated and individuals are severely overlapping, this embodiment adopts a density map estimation algorithm instead of the YOLOv8 detection method, and loads a regression-based convolutional neural network (such as CSRNet or MCNN) model. This model is trained using the density map dataset pre-made in this invention (that is, in the training images, each manually labeled wild horse head position is rendered as a two-dimensional Gaussian kernel).

[0076] 4.1.2) Input distant view or aerial data from video or image data streams into a regression-based convolutional neural network model, and output a density heatmap. The value of each pixel in the density heatmap represents the probability density of the presence of an animal at that location.

[0077] 4.1.3) Integrate and sum all pixel values ​​in the density heatmap. This sum is the estimated total number in the distant view or aerial data, and then outputs a "time-quantity" data pair (e.g., [2023-10-28 08:00:00, 128 plc]). This method is more accurate in dense scenes than bounding box-based counting.

[0078] 4.2) Perform data aggregation and rhythm analysis on the "time-quantity" data, and output the activity rhythm diagram of the animal group and related statistical reports: 4.2.1) Call the cyclic statistics library (such as Python's pycircstat library or R's circular package).

[0079] 4.2.2) Group the "time-quantity" data according to the preset time unit and calculate the average number of animals per hour.

[0080] 4.2.3) Treat 24 hours as a circle, use the average number of animals per hour as the vector size at that time point, and calculate the mean vector of the data for the whole day. The direction of the vector indicates the peak time of activity, and its length (R value) indicates the significance of the rhythm (the closer the value is to 1, the more obvious the rhythm).

[0081] 4.2.4) Output the activity rhythm diagram of the animal group and related statistical reports (such as peak time and rhythm intensity).

[0082] 5) Output refined individual analysis results and macro-group analysis results of Przewalski's horses to solve the problems of individual tracking under complex occlusion and group dynamic analysis in large-scale scenes.

[0083] Example 3 This embodiment provides a processing device corresponding to the Przewalski's horse recognition method based on video tracking provided in Embodiment 2. The processing device can be applied to client processing devices, such as mobile phones, laptops, tablets, desktop computers, etc., to execute the method of Embodiment 2.

[0084] The processing device includes a processor, a memory, a communication interface, and a bus. The processor, memory, and communication interface are connected via the bus to enable communication between them. The memory stores a computer program that can run on the processing device. When the processing device runs the computer program, it executes the Przewalski's horse recognition method based on video tracking provided in Embodiment 2.

[0085] In some implementations, the memory may be high-speed random access memory (RAM), and may also include non-volatile memory, such as at least one disk storage device.

[0086] In other implementations, the processor can be any type of general-purpose processor, such as a central processing unit (CPU) or a digital signal processor (DSP), and there is no limitation here.

[0087] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0088] Those skilled in the art will understand that the structure of the above-described computing device is only a partial structure related to the present invention and does not constitute a limitation on the computing device to which the present invention is applied. A specific computing device may include more or fewer components, or combine certain components, or have different component arrangements.

[0089] Example 4 This embodiment provides a computer program product corresponding to the Przewalski's horse identification method based on video tracking provided in Embodiment 2. The computer program product may include a computer-readable storage medium on which computer-readable program instructions for executing the Przewalski's horse identification method based on video tracking described in Embodiment 2 are loaded.

[0090] A computer-readable storage medium can be a tangible device that holds and stores instructions for use by an instruction execution device. A computer-readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination thereof.

[0091] The computer-readable storage medium provided in the above embodiments has a similar implementation principle and technical effect to the above method embodiments, and will not be described again here.

[0092] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0093] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0094] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0095] The above embodiments are only used to illustrate the present invention. The structure, connection method and manufacturing process of each component can be varied. All equivalent transformations and improvements made on the basis of the technical solution of the present invention should not be excluded from the protection scope of the present invention.

Claims

1. A Przewalski's horse recognition system based on video tracking, characterized in that, include: A data input interface for receiving video or image data streams from one or more camera devices; The data type determination module is used to determine the data type of a video or image data stream based on its metadata. The refined individual analysis channel is used to accurately track, optimize trajectories, and analyze behavior of Przewalski's horses in close-up data from video or image data streams, resulting in refined individual analysis results of Przewalski's horses. The macro-population analysis channel is used to perform population statistics and rhythm analysis on distant or aerial data in video or image data streams to obtain macro-population analysis results of Przewalski's horses; The data output interface is used to output refined individual analysis results and macro-group analysis results of Przewalski's horses to solve the problems of individual tracking under complex occlusion and group dynamic analysis in large-scale scenes.

2. The Przewalski's horse recognition system based on video tracking as described in claim 1, characterized in that, The refined individual analysis channel includes: The target detection module is used to use a pre-trained target detection model to detect all Przewalski's horse individuals in the video frames of the video or image data stream sent by the data input interface, and output the bounding box coordinates of each Przewalski's horse individual target. The multi-target tracking module is used to predict the motion state and associate data of individual Przewalski's horses based on the bounding box coordinates of the individual Przewalski's horses using a multi-target tracking algorithm. It assigns a temporary identity to each individual Przewalski's horse that appears in a video frame and outputs the original tracking trajectory of each individual Przewalski's horse. The trajectory optimization module is used to determine whether each Przewalski's horse individual target has experienced an occlusion loss event based on the original tracking trajectory of each Przewalski's horse individual target, and to obtain the optimized tracking trajectory of each Przewalski's horse individual target by using an occlusion-location association matching method. The individual behavior analysis module is used to classify the behavior of each individual Przewalski's horse based on the optimized tracking trajectory and corresponding video clips, and on the kinematic and / or spatiotemporal video features of the individual Przewalski's horse. The module outputs the duration and time interval of each individual Przewalski's horse performing a certain behavior.

3. The Przewalski's horse recognition system based on video tracking as described in claim 2, characterized in that, The trajectory optimization module includes: The trajectory and occlusion monitoring unit is used to monitor the trajectory status of all Przewalski's horse individual targets in each video frame in real time and calculate the degree of overlap between the bounding boxes of each Przewalski's horse individual target. The occlusion loss event determination unit is used to determine whether an occlusion loss event has occurred for each Przewalski's horse individual target based on the degree of overlap between the bounding boxes of each Przewalski's horse individual target and a pre-set overlap threshold. The occlusion context recording unit is used to mark the identity of the Przewalski's horse individual target that has experienced an occlusion loss event as a waiting reconnection state, and at the same time record the identity of the Przewalski's horse individual target associated with the Przewalski's horse individual target that has experienced an occlusion loss event and its current original tracking trajectory, and use the associated Przewalski's horse individual target as the associated occlusion object of the Przewalski's horse individual target that has experienced an occlusion loss event. The new trajectory discovery unit is used to assign a new identity to a new Przewalski's horse individual target when a new Przewalski's horse individual target appears that cannot be matched with any original tracking trajectory; The occlusion-location association matching unit is used to calculate the initial position of the new Przewalski's horse individual target based on the data recorded by the occlusion context recording unit when the new Przewalski's horse individual target appears, and to calculate the spatial distance between the target and the current position of the occluding object individual, wherein the occluding object individual is the Przewalski's horse individual target associated with the Przewalski's horse individual target that has experienced an occlusion loss event; The identity reconnection unit is used to determine whether the new Przewalski's horse individual target is an occluded individual based on the calculated spatial distance and a preset spatial distance threshold. If so, the identity reconnection operation is performed to correct the identity of the new Przewalski's horse individual target to that of the occluded individual, clear the waiting reconnection state of the occluded individual, and merge the two tracking trajectories into one optimized tracking trajectory. The occluded individual is the Przewalski's horse individual target that has experienced an occlusion loss event.

4. The Przewalski's horse recognition system based on video tracking as described in claim 1, characterized in that, The macro-group analysis channel includes: The population statistics module is used to automatically detect and count the total number of Przewalski's horses in distant views or aerial data in video or image data streams using target detection models or density map estimation algorithms, and output "time-count" data pairs. The group rhythm analysis module is used to aggregate and analyze the rhythm of "time-quantity" data pairs, and output the activity rhythm diagram of the animal group and related statistical reports.

5. A Przewalski's horse identification method based on video tracking, characterized in that, include: Receive video or image data streams from one or more camera devices; Determine the data type of the video or image data stream based on its metadata. When the data type of the video or image data stream is close-up data, the Przewalski's horse individual is accurately tracked, its trajectory optimized, and its behavior analyzed to obtain refined individual analysis results of the Przewalski's horse. When the data type of the video or image data stream is a distant view or aerial photography data, population statistics and rhythm analysis are performed on the video or image data stream to obtain the macroscopic population analysis results of Przewalski's horses; It outputs refined individual analysis results and macro-group analysis results of Przewalski's horses to solve the problems of individual tracking under complex occlusion and group dynamic analysis in large-scale scenes.

6. The Przewalski's horse identification method based on video tracking as described in claim 5, characterized in that, When the data type of the video or image data stream is close-up data, the video or image data stream is used for precise tracking, trajectory optimization, and behavior analysis of individual Przewalski's horses to obtain refined individual analysis results of Przewalski's horses, including: A pre-trained target detection model is used to detect all Przewalski's horse individuals in video frames of video or image data streams, and output the bounding box coordinates of each Przewalski's horse individual target. A multi-target tracking algorithm is adopted. Based on the bounding box coordinates of individual Przewalski's horse targets, the motion state of individual Przewalski's horse targets is predicted and data is associated. A temporary identity is assigned to each individual Przewalski's horse target that appears in the video frame, and the original tracking trajectory of each individual Przewalski's horse target is output. Based on the original tracking trajectory of each Przewalski's horse individual target, it is determined whether each Przewalski's horse individual target has experienced an occlusion loss event, and the occlusion-location association matching method is used to obtain the optimized tracking trajectory of each Przewalski's horse individual target. Based on the optimized tracking trajectory of each Przewalski's horse individual target and the corresponding video or image data stream, the behavior of each individual target is classified according to the kinematic characteristics and / or spatiotemporal video characteristics, and the duration and time interval of each individual Przewalski's horse performing a certain behavior are output.

7. The Przewalski's horse identification method based on video tracking as described in claim 6, characterized in that, The process involves determining whether an occlusion loss event has occurred for each Przewalski's horse individual target based on its original tracking trajectory, and then using an occlusion-location association matching method to obtain an optimized tracking trajectory for each individual Przewalski's horse target, including: The trajectory status of all Przewalski's horse individual targets in each video frame is monitored in real time, and the degree of overlap between the bounding boxes of each Przewalski's horse individual target is calculated. Based on the degree of overlap between the bounding boxes of each Przewalski's horse individual target and a pre-set overlap threshold, it is determined whether each Przewalski's horse individual target has experienced an occlusion loss event. When an occlusion loss event is triggered, the identity of the Przewalski's horse individual target that experienced the occlusion loss event is marked as waiting for reconnection and added to the waiting reconnection list. At the same time, the identity of the Przewalski's horse individual target associated with the Przewalski's horse individual target that experienced the occlusion loss event and its current original tracking trajectory are recorded, and the associated Przewalski's horse individual target is used as the associated occlusion object of the Przewalski's horse individual target that experienced the occlusion loss event. Monitor whether any new Przewalski's horse individual target tracks are initialized. When a new Przewalski's horse individual target exists, assign a new identity to the new Przewalski's horse individual target and trigger occlusion-location association matching. Based on the waiting reconnection list and the identity of the Przewalski's horse individual targets associated with the occlusion loss event and their current original tracking trajectories, occlusion-location association matching is performed. Based on the calculated spatial distance and the preset spatial distance threshold, it is determined whether the new individual is an occluded individual. If so, an identity reconnection operation is performed to correct the identity of the new individual to that of the occluded individual, clear its waiting reconnection state, and merge the two tracking trajectories into one optimized tracking trajectory.

8. The Przewalski's horse identification method based on video tracking as described in claim 5, characterized in that, When the data type of the video or image data stream is a distant view or aerial photography data, population statistics and rhythm analysis are performed on the video or image data stream to obtain macroscopic population analysis results of Przewalski's horses, including: Using target detection models or density map estimation algorithms, the total number of Przewalski's horses in distant views or aerial data in video or image data streams is automatically detected and counted, and "time-count" data pairs are output. Data aggregation and rhythm analysis are performed on the "time-quantity" data to output the activity rhythm diagram of the animal group and related statistical reports.

9. A processing device, characterized in that, It includes computer program instructions, wherein when executed by a processing device, the computer program instructions are used to implement the steps corresponding to the video tracking-based Przewalski's horse recognition method according to any one of claims 5-8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions, wherein the computer program instructions, when executed by a processor, are used to implement the steps corresponding to the video tracking-based Przewalski's horse recognition method according to any one of claims 5-8.

Images