Similar wild protection animal interference data processing method and system

Abstract

The invention discloses a similar wild protected animal interference data processing method and system, and belongs to the field of computer vision, and the method comprises the steps: obtaining one or more wild protected animal monitoring reference data sets, and constructing a data space D; m single-target tracking algorithms are selected to construct a determiner set A; performing similar object interference feature extraction and judgment on each video sequence in the data space D to obtain a frame-level judgment result sequence and sequence-level interference statistical data; and performing grading processing according to the sequence-level interference statistical data to obtain a structured data set containing similar object interference characteristics. Aiming at the problem of tracking failure caused by similar object interference in a wild animal tracking technology in the prior art, the invention provides an interference identification mechanism based on multi-determiner cooperation, and automatic detection, quantitative analysis and grading processing of similar object interference in a video sequence are realized.

Description

Technical Field

[0001] This application relates to the field of computer vision, and more specifically, to a method and system for processing data related to interference from similar wildlife conservation animals. Background Technology

[0002] Wildlife conservation is an important part of global ecological environment protection. With the development of computer vision technology, intelligent wildlife tracking systems based on video surveillance have gradually become a research hotspot, and a large number of wildlife conservation monitoring video datasets have been established for algorithm development and evaluation.

[0003] However, wildlife video tracking faces unique technical challenges, particularly the problem of interference from similar objects: In the natural environment, different individuals of the same species are highly similar in terms of body size, fur color, and behavioral patterns, such as in giant panda identification; closely related species with similar morphological characteristics often appear in the wild, such as leopards, snow leopards, and clouded leopards, which share similar body size and marking patterns; the camouflage and seasonal appearance changes of wild animals further exacerbate the difficulty of tracking. Existing general-purpose single-target tracking algorithms perform poorly when dealing with these wildlife-specific similarity interferences.

[0004] While a large number of wildlife monitoring video datasets exist, these datasets lack systematic identification, annotation, and classification of interference phenomena caused by similar objects. Existing datasets primarily rely on manual annotation, which presents the following problems:

[0005] First, manually identifying interference from similar objects is extremely costly and inefficient. It requires hiring a large number of annotators to watch massive amounts of video frame by frame and relying on manual judgment to identify and label segments with interference from similar objects. This process is time-consuming, labor-intensive, and costly.

[0006] Second, manual judgment is highly subjective and lacks consistency. The criteria for judging similarity vary significantly among different annotators, making it difficult to ensure consistent annotation quality and establish an objective and unified evaluation standard.

[0007] Third, there is a lack of quantitative analysis and grading of interference from similar objects. Existing methods cannot accurately count key indicators such as the frequency and duration of interference, nor can they perform structured grading of data according to the severity of interference, making it difficult to provide targeted data support for different application scenarios.

[0008] Fourth, there is a bias in human-computer cognition. The human visual system's perception of similarity differs from that of computer vision algorithms. Data sets built based on human subjective judgment may not accurately reflect the scenarios that pose a challenge to the algorithms.

[0009] Therefore, existing technologies lack effective methods for automatic identification and classification of interference from similar objects in wildlife video monitoring data. There is an urgent need for a data processing method that can automatically and objectively identify interference from similar objects and perform quantitative analysis and structured classification of the data. Summary of the Invention

[0010] To address the problem of tracking failure caused by interference from similar objects in existing wildlife tracking technologies, this application provides a method and system for processing interference data from similar wildlife. By constructing a multi-determiner collaborative interference identification mechanism, it achieves automatic detection, quantitative analysis, and graded processing of interference from similar objects in video sequences.

[0011] One aspect of this application provides a method for processing data on interference from similar wild animals, comprising: S1, acquiring one or more benchmark datasets for monitoring wild animals and constructing a data space D; S2, selecting m single-target tracking algorithms to construct a set of determiners A; S3, extracting and determining interference features of similar objects for each video sequence in the data space D, obtaining a frame-level determination result sequence and a sequence-level interference statistics data; S4, performing hierarchical processing based on the sequence-level interference statistics data to obtain a structured dataset containing interference features of similar objects.

[0012] In this scheme, the single-target tracking algorithm serves as the foundation for the decision-maker. By providing the initial position of the target in the first frame, it can automatically locate and track the target in subsequent frames. These algorithms generate confidence score maps to represent the probability of each location within the search area being a target object, thereby enabling continuous tracking of protected wild animals.

[0013] The decider is an interference detection tool that specifically optimizes the original single-target tracking algorithm for the biological characteristics (such as movement patterns and appearance features) of wild protected animals. Each decider can independently analyze video frames and determine the presence of similar objects as interference by identifying the number of candidate targets. Multiple deciders form a decider set A = {a1, a2, ..., a...} m By working collaboratively through a voting mechanism, the accuracy and robustness of interference detection can be improved.

[0014] Furthermore, S2, select m single-target tracking algorithms to construct a decision set A, including: selecting m candidate tracking algorithms from a preset monocular tracking algorithm library based on the biological characteristic parameters of wild protected animals, where m is greater than or equal to 2; configuring parameters for each candidate tracking algorithm, including setting target detection threshold, tracking confidence threshold and similarity matching parameters;

[0015] Based on the movement patterns and appearance characteristics of wild protected animals, feature extractors are optimized for each candidate tracking algorithm to generate an adaptive decision maker. Feature extractor optimization includes: adjusting the convolutional layer parameters of the feature extraction network to enhance the extraction ability of animal texture, contour and color features, optimizing the feature vector dimension to improve the distinguishability of similar objects, and configuring multi-scale feature fusion weights.

[0016] The performance of the generated m decision-makers is evaluated, and the tracking accuracy and similar object discrimination index of each decision-maker on the standard test set are calculated.

[0017] Based on the performance evaluation results, weights are assigned to the decision makers, and a decision maker set A = {a1, a2, ..., a...} is constructed. m}

[0018] Among them, biological characteristic parameters of wild protected animals refer to a set of parameter indicators used to describe and quantify the species characteristics of wild protected animals. In this application, these include: morphological parameters, appearance characteristic parameters, locomotion parameters, etc.

[0019] Further, in step S3, similar object interference features are extracted and determined for each video sequence in the data space D, resulting in a frame-level determination result sequence and sequence-level interference statistics. This includes: for each video sequence in the data space D, calculating the confidence score map for each frame using each determiner in the determiner set A, where the confidence score map represents the confidence level that each position within the search area is a target object; performing local peak processing on each confidence score map, extracting local maxima through max pooling to generate local maximum confidence; selecting candidate targets whose local maximum confidence is greater than a preset threshold to form a candidate target set; and performing a determination on each video sequence in the data space D. The system determines interference based on the number of candidates in the candidate target set: if the number of candidates is greater than 1, the corresponding frame is determined to have interference from similar objects; otherwise, it is determined that there is no interference from similar objects. A voting mechanism is used to summarize the determination results of all the determiners and count the number of votes in favor. When the number of votes in favor exceeds a preset threshold, it is confirmed that the current frame has interference from similar objects. The system repeats the interference determination for all frames in the video sequence to generate a corresponding frame-level determination result sequence. Based on the frame-level determination result sequence, the system calculates sequence-level interference statistics, including: the number of interfering frames, the proportion of interfering frames, the number of consecutive interfering segments, and the average duration of interference.

[0020] Furthermore, candidate targets with a local maximum confidence score greater than a preset threshold are selected to form a candidate target set, including: obtaining the global maximum value R in the local maximum confidence score map of the current frame. max Set the relative thresholds θ×R respectively. maxAnd the absolute threshold τ; where the relative threshold is dynamically adjusted with the global maximum value of each frame to adapt to the screening of candidate targets in different scenarios; traverse each position in the local maximum confidence score map, and mark the positions whose confidence values ​​are both greater than the relative threshold and the absolute threshold as valid candidate targets; form a candidate target set by combining all the positions marked as valid candidate targets and their corresponding confidence values.

[0021] In particular, wildlife videos are filmed in complex and variable environments with vastly different lighting conditions and background interference, resulting in drastic variations in the absolute range of confidence scores across different scenarios. A single fixed threshold cannot adapt to such dynamic changes. Similar objects often produce the second-highest confidence peaks, requiring a mechanism that can both capture these secondary peaks and filter out noise interference.

[0022] This application achieves dynamic normalization of the confidence evaluation criterion by associating the threshold with the global maximum value Rmax of the current frame. This allows the algorithm to automatically adapt to differences in confidence distribution under different scenarios without manual parameter tuning. The parameter θ, θ∈(0,1) controls the sensitivity to similar objects. When there are interference objects that are highly similar to the target, their confidence is usually a certain proportion of the main target's. By adjusting θ, the sensitivity of this similarity detection can be precisely controlled.

[0023] In challenging scenarios where target confidence is generally low, a purely relative threshold may misclassify a large amount of background noise as candidate targets. An absolute threshold provides a safety net, ensuring that only locations with a certain level of confidence are considered. This prevents the explosion of false detections that can occur with relative thresholding mechanisms in cases of target loss or severe occlusion.

[0024] Furthermore, a voting mechanism is used to summarize the judgment results of all determiners, count the number of affirmative votes, and when the number of affirmative votes exceeds a preset threshold, it is confirmed that there is similar object interference in the current frame. This includes: for each determiner in the determiner set A, judging based on the number of candidate targets in the candidate target set: if the number of candidate targets in the candidate target set is greater than 1, the corresponding determiner's judgment result for the current frame is set to 1, indicating that there is similar object interference; if the number of candidate targets in the candidate target set is less than or equal to 1, the corresponding determiner's judgment result for the current frame is set to 0, indicating that there is no similar object interference; the judgment results of all determiners in the determiner set A are counted, and the number of determiners with a judgment result of 1 is calculated as the total number of affirmative votes obtained in the current frame; a judgment threshold β is set, and when the total number of affirmative votes is greater than the judgment threshold β, it is confirmed that there is similar object interference in the current frame, otherwise it is confirmed that there is no similar object interference in the current frame; the final similar object interference judgment result for the current frame is output.

[0025] Furthermore, based on the frame-level determination result sequence, sequence-level interference statistics are calculated, including: performing similar object interference determination on each frame in the video sequence in chronological order, obtaining the final similar object interference determination result for each frame; and assembling the determination results of all frames into a frame-level determination result sequence F = {f1, f2, ..., f...} according to the frame number. n}, where n is the total number of frames in the video sequence, f i =1 indicates that there is interference from similar objects in the i-th frame, f i =0 indicates that there is no similar object interference in the i-th frame; according to the frame-level judgment result sequence F, calculate the number of interference frames, count the total number of frames with a judgment result of 1 in the sequence, and calculate the proportion of interference frames; traverse the frame-level judgment result sequence F, mark the consecutive frame sequences with a judgment result of 1 as an interference segment, count the total number of all interference segments as the number of consecutive interference segments; count the number of frames in each consecutive interference segment, calculate the average number of frames in all interference segments as the average interference duration; combine the number of interference frames, the proportion of interference frames, the number of consecutive interference segments, and the average interference duration to obtain the sequence-level interference statistics.

[0026] Furthermore, in step S4, the sequence-level interference statistics are processed hierarchically to obtain a structured dataset containing interference features of similar objects. This includes: merging temporally adjacent interference frames into a single interference event based on the frame-level judgment result sequence F; counting the total number of interference events in each video sequence; and classifying the video sequences according to the total number of interference events: when the total number of interference events equals 1, the corresponding video sequence is divided into a single interference subset; single interference often corresponds to sporadic scenarios, such as accidental encounters between animals; when the total number of interference events is greater than 1 but less than a preset interference threshold, the corresponding video sequence is divided into a multiple interference subset; multiple interference reflects periodic interference patterns, such as repeated contact caused by overlapping territories; when the total number of interference events is greater than or equal to a preset interference threshold, the corresponding video sequence is divided into a frequent interference subset; frequent interference suggests continuous complex scenarios, such as continuous interaction among social animals; and combining the single interference subset, multiple interference subset, and frequent interference subset to construct a structured dataset containing interference from similar objects.

[0027] Furthermore, based on the frame-level determination result sequence F, temporally adjacent interference frames are merged into a single interference event, including: when the time interval between two adjacent frames with a determination result of 1 is less than M frames, they are determined to be the same continuous interference event; when the time interval between two adjacent frames with a determination result of 1 is greater than or equal to M frames, they are determined to be two independent interference events.

[0028] In particular, interference from similar objects in wildlife is often a continuous process, such as the interlacing movements of two similar individuals. Simple frame-level annotation disrupts this temporal continuity, while event-level representation restores the complete semantics of the interference phenomenon. By setting the time interval M, it is possible to tolerate brief misjudgments or omissions during the judgment process, avoiding the fragmentation of a continuous interference event into multiple events due to the judgment error of individual frames, thus improving the robustness of the system.

[0029] Another aspect of this application provides a similar wildlife interference data processing system, comprising: a data acquisition module for acquiring one or more wildlife monitoring benchmark data and constructing a data space D; a decision-maker module for selecting m single-target tracking algorithms to construct a decision-maker set A; an interference determination module for extracting similar object interference features from each video sequence in the data space D to obtain a frame-level determination result sequence and a sequence-level interference statistics data; and a hierarchical processing module for performing hierarchical processing on the video sequences based on the sequence-level interference statistics data to obtain a structured dataset containing similar object interference features.

[0030] Furthermore, the interference determination module includes: a confidence calculation unit, which uses each determiner in the determiner set A to calculate the confidence score map for each frame of each video sequence in the data space D, where the confidence score map represents the confidence level that each position within the search area is a target object; a local peak processing unit, which performs local peak processing on each confidence score map, extracts the local maximum value through max pooling, and generates the local maximum value confidence; a candidate target unit, which selects candidate targets whose local maximum value confidence is greater than a preset threshold to form a candidate target set; and a single frame determination unit, which, for each determiner, determines the number of candidate objects in the candidate target set. The system performs several checks: 1) A decision-making unit uses a voting mechanism to summarize the results of all checkers, counting the number of votes in favor. When the number of votes exceeds a preset threshold, it confirms that the current frame has interference from similar objects. 2) A statistical analysis unit performs repeated interference checks on all frames in the video sequence, generating a corresponding frame-level decision result sequence. Based on the frame-level decision result sequence, it calculates sequence-level interference statistics, including: the number of interfering frames, the proportion of interfering frames, the number of consecutive interfering segments, and the average duration of interference.

[0031] Compared to existing technologies, the advantages of this application are:

[0032] This application first constructs a data space from a benchmark dataset for monitoring protected wild animals, and selects multiple single-target tracking algorithms to build a set of decision makers. Then, it performs multi-level similar object interference feature extraction on each video sequence, including confidence score map calculation, local peak processing, and candidate target screening. The frame-level interference judgment result is determined through a multi-decision maker voting mechanism. Then, it statistically analyzes sequence-level interference feature data, including the number of interference frames, the proportion of interference frames, the number of consecutive interference segments, and the average duration of interference. Finally, it classifies the video sequences according to the frequency of interference events, and constructs a structured dataset containing three levels: single interference, multiple interference, and frequent interference.

[0033] This application aims to automatically identify interference from similar objects in wildlife conservation monitoring videos, thereby improving the accuracy and robustness of interference detection; and to construct a hierarchical interference dataset to provide targeted test and training data for different application scenarios. Attached Figure Description

[0034] Figure 1 This is a flowchart illustrating an embodiment of the data mining method for similar object interference challenges in this application;

[0035] Figure 2 This is a schematic diagram illustrating the structure of an embodiment of the similar object interference challenge data mining system of this application. Detailed Implementation

[0036] The present application will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0037] Examples of the similar object interference challenge mining method of this application, such as Figure 1 As shown, it includes the following steps:

[0038] In step S10, representative general single-target tracking benchmark environments are combined to form a data space to be mined: D = {D1, D2, D3, ..., D...} d}

[0039] In step S20, representative single-target tracking algorithms are combined to form a set of similar object interference challenge determiners A = {A1, A2, A3, ..., A...} to be run. m}

[0040] In step S30, video sequences are extracted from the data space D, the tracking algorithm in the decision set A is run, and the sequence selection is performed using a similar object interference challenge data mining method. This method includes two steps: frame-level decision-making and sequence-level selection.

[0041] Frame-level determination: First, perform max pooling on the confidence score map to extract local maxima:

[0042] R local =MaxPool(R) (1)

[0043] Then, further filtering of the local maximum confidence score map, the maximum confidence score R is selected. max As a reference value to remove distractors with low confidence scores:

[0044] C = (i,j) / R local (i,j)=R(i,j);

[0045] R local (i,j)≥θ×R max ;

[0046] R local (i,j)≥τ; (2)

[0047] Where C is the set of target candidates after screening, and R max θ represents the global maximum value of the confidence score in the current frame, θ is the confidence score threshold hyperparameter, and τ is the absolute confidence threshold hyperparameter.

[0048] The results of all trackers are collected, and a voting mechanism is used to make the final judgment on the interference challenge of similar objects;

[0049] In step S40, based on the screening results of the similar object interference challenge video sequences in step S30, the mined similar object interference challenge dataset is constructed.

[0050] In this embodiment, the video sequences are further divided according to the frequency of similar object interference challenges to obtain subsets with different levels of interference.

[0051] In this embodiment, the frequency of occurrence of the similar object interference challenge is approximately equal to the interference level of the similar object interference challenge.

[0052] The subsets can be divided according to the partitioning criteria as follows:

[0053] Single interference: A similar object appears only once in the video sequence. This interference may be sudden or long-lasting.

[0054] Multiple interferences: Similar objects appear multiple times in the video sequence, but the number of times they appear is lower than the category of frequent interferences.

[0055] Frequent interference: The appearance of similar objects multiple times in a video sequence often indicates that the video scene is relatively complex, and the interference seriously affects the tracking performance.

[0056] In this embodiment, equation (3) is used to classify the three levels of interference:

[0057] DOnce ={L i |T i =1};

[0058] D more ={L i |T i ∈(1,10]};

[0059] D most ={L i |T i >10} (3)

[0060] Among them, D Once D more D most These represent single-time disturbance subsets, multiple-time disturbance subsets, and frequent-time disturbance subsets, respectively.

[0061] In this embodiment, the extracted similar object interference challenge dataset is denoted as D. soi The subset of data has the relationship shown in equation (5).

[0062] Department: D soi ={D Once ∪D more ∪D most}(5)

[0063] Another aspect of this application proposes an intelligent similar object interference challenge data mining system, based on the above-described intelligent similar object interference challenge data mining method; such as Figure 2 As shown, it includes: a video frame similar object interference challenge judgment unit 11, a video sequence similar object interference challenge statistics unit 12, a video sequence filtering unit 13, and a dataset fine-grained partitioning unit 14.

[0064] The video frame similar object interference challenge determination unit 11 is used to determine whether there are multiple target candidates in the video frame based on the prediction confidence score map result of the tracking algorithm, that is, whether there is a similar object interference challenge, and obtain the video frame level determination result.

[0065] The video sequence similar object interference challenge statistics unit 12 is used to determine whether there is a similar object interference challenge in the video sequence based on the video frame-level statistical results determined by the video frame similar object interference challenge determination unit.

[0066] The video sequence filtering unit 13, based on the similar object interference challenge video sequence judgment results obtained by the video sequence similar object interference challenge statistics unit, mines similar object interference challenge video sequences to form a similar object interference challenge dataset.

[0067] The dataset fine-grained partitioning unit 14 is used to perform fine-grained partitioning of similar object interference challenge video sequences based on the interference degree level of similar object interference challenge, so as to obtain similar object interference challenge subset datasets.

[0068] Furthermore, data mining systems also include:

[0069] Evaluation and assessment unit 15 is used to evaluate and analyze the algorithm on the mined dataset.

[0070] Those skilled in the art will recognize that the method steps, units, and subsystems of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of electronic hardware and software, the components and steps of each example have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0071] The technical solutions of this application have been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of this application is obviously not limited to these specific embodiments. Without departing from the principles of this application, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of this application.

Claims

1. A method for processing similar wildlife disturbance data, characterized by, Comprising: S1, obtaining one or more wild animal monitoring benchmark data sets, constructing a data space D; S2, selecting m single-target tracking algorithms to construct a decision maker set A; S3, performing similar object interference feature extraction and decision for each video sequence in the data space D, obtaining a frame-level decision result sequence and sequence-level interference statistical data; S4, performing hierarchical processing according to the sequence-level interference statistical data, obtaining a structured data set containing similar object interference features.

2. The similar wildlife interference data processing method according to claim 1, characterized in that: S2, selecting m single-target tracking algorithms to construct a decision maker set A, comprising: From the preset monocular tracking algorithm library, according to the biological characteristic parameters of the wildlife, selecting m candidate tracking algorithms, wherein m is greater than or equal to 2; Performing parameter configuration on each candidate tracking algorithm, the parameter configuration including setting target detection threshold, tracking confidence threshold and similarity matching parameters; According to the motion mode and appearance features of the wildlife, optimizing the feature extractors of each candidate tracking algorithm to generate adaptive decision makers; Performing performance evaluation on the generated m decision makers, calculating the tracking accuracy and similar object distinguishing index of each decision maker on the standard test set. According to the performance evaluation result, weight distribution is performed on the determinators to construct a determinator set .

3. The similar wildlife interference data processing method according to claim 1, characterized in that: S3, obtaining a frame-level decision result sequence and sequence-level interference statistical data, comprising: For each video sequence in the data space D, using each decision maker in the decision maker set A to calculate a confidence score map for each frame, the confidence score map representing the confidence degree of each position in the search area as a target object; Performing local peak value processing on each confidence score map, extracting local maximum values by maximum pooling operation to generate local maximum confidence; Selecting candidate targets with local maximum confidence greater than a preset threshold to form a candidate target set; For each decision maker, according to the number of candidate objects in the candidate target set, when the number of candidate objects is greater than 1, it is determined that there is similar object interference in the corresponding frame, otherwise it is determined that there is no similar object interference; Using a voting mechanism to aggregate the decision results of all decision makers, and when the number of votes exceeds a preset threshold, it is confirmed that there is similar object interference in the current frame; Repeating the similar object interference decision for all frames in the video sequence to generate a frame-level decision result sequence corresponding to the sequence; According to the frame-level decision result sequence of the sequence, calculating sequence-level interference statistical data, including: interference frame number, interference frame proportion, continuous interference segment number and average interference duration.

4. The similar wildlife interference data processing method according to claim 3, characterized in that: Forming a candidate target set, comprising: obtaining a global maximum value in the current frame local maximum value confidence score map ; Set relative thresholds respectively And the absolute threshold τ; where the relative threshold is dynamically adjusted with the global maximum value of each frame to adapt to the selection of candidate targets in different scenarios; Traversing each position in the local maximum confidence score map, marking positions with confidence values greater than a relative threshold and greater than an absolute threshold as valid candidate targets; Forming a candidate target set by marking all positions marked as valid candidate targets and the corresponding confidence values.

5. The similar wildlife interference data processing method according to claim 3, characterized in that: Confirming that the current frame has similar object interference, comprising: For each decision maker in the decision maker set A, the number of target candidates in the candidate target set is determined: If the number of candidate targets in the candidate target set is greater than 1, the decision result of the corresponding decision maker for the current frame is set to 1, indicating that there is similar object interference; If the number of candidate targets in the candidate target set is less than or equal to 1, the decision result of the corresponding decision maker for the current frame is set to 0, indicating that there is no similar object interference; Statistical results of all decision makers in the decision maker set A are calculated, and the number of decision makers with decision result of 1 is calculated as the total number of votes obtained by the current frame; Set the decision threshold β, when the total number of votes is greater than the decision threshold β, it is confirmed that the current frame has similar object interference, otherwise it is confirmed that the current frame has no similar object interference; Output the final similar object interference decision result of the current frame.

6. The similar wildlife interference data processing method according to claim 3, characterized in that: Calculating sequence-level interference statistical data, comprising: For each frame in the video sequence, the similar object interference determination is performed in time sequence to obtain the final similar object interference determination result of each frame; The determination results of all frames are combined according to frame serial numbers to form a frame-level determination result sequence wherein n is the total number of frames of the video sequence, represents that the i-th frame exists similar object interference, represents that the i-th frame does not exist similar object interference; According to the frame-level determination result sequence F, the number of interference frames is calculated, the total number of frames with determination result of 1 in the sequence is calculated, and the interference frame proportion is calculated; Traverse the frame-level determination result sequence F, mark the frame sequence with continuous determination result of 1 as an interference segment, and calculate the number of all interference segments as the number of continuous interference segments; Statistical results of all decision makers in the decision maker set A are calculated, and the number of decision makers with decision result of 1 is calculated as the total number of votes obtained by the current frame; Set the decision threshold β, when the total number of votes is greater than the decision threshold β, it is confirmed that the current frame has similar object interference, otherwise it is confirmed that the current frame has no similar object interference; Output the final similar object interference decision result of the current frame.

7. The similar wildlife interference data processing method according to claim 1, characterized in that: S4, obtaining a structured data set containing similar object interference features, comprising: Based on the frame-level determination result sequence F, adjacent interference frames in time sequence are merged into a single interference event; Statistical results of all decision makers in the decision maker set A are calculated, and the number of decision makers with decision result of 1 is calculated as the total number of votes obtained by the current frame; Set the decision threshold β, when the total number of votes is greater than the decision threshold β, it is confirmed that the current frame has similar object interference, otherwise it is confirmed that the current frame has no similar object interference; Output the final similar object interference decision result of the current frame.

8. The similar wildlife interference data processing method according to claim 7, characterized in that: Based on the frame-level determination result sequence F, adjacent interference frames in time sequence are merged into a single interference event, comprising: When the time interval between the two adjacent frames with determination result of 1 is less than M frames, it is determined as the same continuous interference event; When the time interval between the two adjacent frames with determination result of 1 is greater than or equal to M frames, it is determined as two independent interference events. Comprising: ​ 9. A similar wildlife disturbance data processing system characterized by, ​ a data acquisition module configured to acquire one or more wildlife monitoring benchmark data and construct a data space D; a determiner module configured to select m single-target tracking algorithms to construct a determiner set A; an interference determination module configured to extract similar object interference features of each video sequence in the data space D to obtain a frame-level determination result sequence and sequence-level interference statistical data; a hierarchical processing module configured to perform hierarchical processing according to the sequence-level interference statistical data to obtain a structured dataset containing similar object interference features.

10. The similar wildlife interference data processing system according to claim 9, wherein the interference determination module comprises: a confidence score calculation unit configured to calculate a confidence score map of each frame of each video sequence in the data space D using each determiner in the determiner set A, wherein the confidence score map represents a confidence degree of each position in a search region as a target object; a local peak processing unit configured to perform local peak processing on each confidence score map, extract a local maximum value through a max-pooling operation, and generate a local maximum value confidence; a candidate target unit configured to select candidate targets with a local maximum value confidence greater than a preset threshold to form a candidate target set; a single-frame determination unit configured to determine, for each determiner, whether there is similar object interference in a corresponding frame according to a number of candidate objects in the candidate target set, wherein when the number of candidate objects is greater than 1, it is determined that there is similar object interference in the corresponding frame, otherwise it is determined that there is no similar object interference; a voting decision unit configured to use a voting mechanism to aggregate determination results of all determiners, and count a number of votes in favor, wherein when the number of votes in favor exceeds a preset threshold, it is confirmed that there is similar object interference in the current frame; a statistical analysis unit configured to repeat similar object interference determination for all frames in a video sequence to generate a frame-level determination result sequence corresponding to the sequence, and calculate sequence-level interference statistical data according to the frame-level determination result sequence, including: a number of interference frames, a proportion of interference frames, a number of continuous interference segments, and an average interference duration. ​

Images