A target tracking method and system combined with multi-channel fusion

Abstract

This invention relates to a target tracking method and system combining multi-channel fusion, belonging to the field of target tracking. It obtains a motion profile of the tracked target, including scene type and target type; performs correlation analysis to obtain a set of motion-sensitive elements; obtains a temporal matrix of sensitive elements in the set; configures channels through confidence constraint regions to retrieve concentrated motion prediction trajectories of historical sample groups that satisfy the motion profile and the temporal matrix of sensitive elements, and statistically analyzes the temporal information of the confidence constraint regions; processes the target motion trajectory monitoring temporal information and a preset element feature value matrix through a motion trajectory prediction channel bound to the motion profile to generate a motion prediction trajectory; when the motion prediction trajectory satisfies the temporal information of the confidence constraint region, the motion prediction trajectory is output, and target tracking pre-configuration is performed. This solves the technical problem that target tracking fails to effectively consider real-time fluctuations of key control parameters, leading to motion trajectory deviation and poor tracking accuracy.

Description

Technical Field

[0001] This invention relates to the field of target tracking, and in particular to a target tracking method and system that combines multi-channel fusion. Background Technology

[0002] In natural conditions, target detection and tracking are crucial research areas in computer vision. Traditional target tracking methods primarily rely on single visual features or motion models. However, these methods are often susceptible to interference from factors such as lighting variations, occlusion, and noise, leading to decreased tracking accuracy or even tracking failure. To address this issue, researchers have begun exploring target tracking methods that combine multiple features and models. Among these, multi-channel fusion technology is an effective approach. By integrating information from different channels, such as color, texture, shape, and motion, this technology can more comprehensively describe the characteristics of the target, thereby improving tracking accuracy and robustness. However, existing multi-channel fusion target tracking methods still face several challenges. On one hand, effectively selecting and fusing information from different channels is a challenge. Different scenes and target types may require different features and models, necessitating a flexible approach to adapt to these variations. On the other hand, real-time fluctuations in key control parameters can significantly impact tracking results. Existing target tracking methods often ignore these parameter fluctuations, leading to deviations in motion trajectories and decreased tracking accuracy. Summary of the Invention

[0003] This invention addresses the technical problem in existing technologies where target tracking fails to effectively consider real-time fluctuations in key control parameters, leading to trajectory deviations and poor tracking accuracy. It provides a target tracking method and system that combines multi-channel fusion to solve this problem.

[0004] The technical solution of the present invention to solve the above-mentioned technical problems is as follows:

[0005] In a first aspect, the present invention provides a target tracking method combining multi-channel fusion, comprising: obtaining a motion profile of the tracked target, wherein the motion profile includes a scene type and a target type; performing correlation analysis based on the scene type and the target type to obtain a set of motion-sensitive elements; obtaining a temporal matrix of sensitive elements in the set of motion-sensitive elements; retrieving a set of concentrated motion prediction trajectories of historical sample groups that satisfy the motion profile and the temporal matrix of sensitive elements through a confidence constraint region configuration channel, and statistically analyzing the temporal information of the confidence constraint region; processing the target motion trajectory monitoring temporal information and a preset element feature value matrix through a motion trajectory prediction channel bound to the motion profile to generate a motion prediction trajectory; and outputting the motion prediction trajectory when the motion prediction trajectory satisfies the temporal information of the confidence constraint region, and performing target tracking pre-configuration.

[0006] Secondly, the present invention provides a target tracking system combining multi-channel fusion, the system comprising: a profile acquisition module for obtaining a motion profile of the tracked target, wherein the motion profile includes a scene type and a target type; a correlation analysis module for performing correlation analysis based on the scene type and the target type to obtain a set of motion-sensitive elements; a time-series analysis module for obtaining a time-series matrix of sensitive elements of the set of motion-sensitive elements; a channel configuration module for configuring channels through a confidence constraint region, retrieving concentrated motion prediction trajectories of historical sample groups that satisfy the motion profile and the time-series matrix of sensitive elements, and statistically analyzing the time-series information of the confidence constraint region; a trajectory prediction module for processing the target motion trajectory monitoring time-series information and a preset element feature value matrix through a motion trajectory prediction channel bound to the motion profile, and generating a motion prediction trajectory; and a tracking configuration module for outputting the motion prediction trajectory and performing target tracking pre-configuration when the motion prediction trajectory satisfies the time-series information of the confidence constraint region.

[0007] The beneficial effects of this invention are as follows: by acquiring the motion profile of the tracked target, performing correlation analysis to obtain the set of motion-sensitive elements and its temporal matrix, using historical sample groups to statistically analyze the temporal information of the confidence constraint region, and generating a motion prediction trajectory through the motion trajectory prediction channel bound to the motion profile, the target tracking pre-configuration is finally output and executed when the confidence constraint conditions are met, thereby improving the accuracy and robustness of target tracking. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating a target tracking method combining multi-channel fusion, as provided by the present invention.

[0009] Figure 2 This is a schematic diagram of a target tracking system combining multi-channel fusion, provided by the present invention.

[0010] Figure labeling: Image acquisition module 11, correlation analysis module 12, time series analysis module 13, channel configuration module 14, trajectory prediction module 15, tracking configuration module 16. Detailed Implementation

[0011] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0012] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0013] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0014] Example 1:

[0015] like Figure 1 As shown, this embodiment of the invention provides a target tracking method combining multi-channel fusion, including:

[0016] S10: Obtain a motion profile of the tracked target, wherein the motion profile includes scene type and target type.

[0017] For example, target tracking refers to the process of detecting, identifying, and continuously locating a target of interest in a video sequence or continuous image frames. It involves the analysis and modeling of information such as the target's appearance and motion patterns to achieve trajectory estimation and state updates of the target over continuous time. Target tracking has wide applications in many fields, such as video surveillance, autonomous driving, drone inspection, and intelligent industrial monitoring. To improve tracking accuracy, this solution proposes a target tracking method combining multi-channel fusion.

[0018] Specifically, in the target tracking process, the first step is to acquire a motion profile of the target. A motion profile is a comprehensive description that mainly includes two aspects: scene type and target type. Scene type refers to the environmental background of the target, such as whether it is a city street, a natural environment, or an indoor space. Different scene types often imply different lighting conditions, background complexity, and potential interference factors, all of which affect the effectiveness of target tracking. For example, in a city street scene, special attention needs to be paid to the movement of vehicles and pedestrians, as well as potential occlusion; while in a natural environment, changes in lighting and complex background textures can be major challenges. Target type refers to the specific category of the object being tracked, such as whether it is a person, vehicle, animal, or other object. Different target types have different appearance characteristics and movement patterns. For example, a person's movement trajectory may be affected by factors such as walking speed, changes in direction, and interactive behaviors; while vehicles may have more stable movement patterns and more obvious appearance characteristics, such as shape and color. Therefore, when acquiring a motion profile, it is necessary to comprehensively consider both scene type and target type information to provide a more accurate and comprehensive foundation for subsequent analysis and processing.

[0019] S20: Perform correlation analysis based on the scene type and the target type to obtain a set of motion-sensitive elements.

[0020] Optionally, after obtaining the motion profile of the tracked target, a correlation analysis is performed based on the scene type and target type to obtain a set of motion-sensitive elements. The correlation analysis aims to filter out sensitive elements that are highly related to the current scene and target from a large number of possible elements.

[0021] The correlation analysis is based on a pre-defined set of elements, constructed using past experience and professional knowledge, and includes various factors that may affect target tracking. For example, in an urban street scene, the element set might include road width, traffic flow, and pedestrian density; while in an indoor space, it might include room layout, furniture placement, and lighting conditions. Next, correlation analysis is performed on these pre-defined elements according to the current scene type and target type, considering the correlation between the elements and the target's motion trajectory, as well as the importance of the elements in different scenes. Through calculation and analysis, sensitive elements highly correlated with the current motion profile can be extracted from the pre-defined element set, forming a motion-sensitive element set.

[0022] For example, in a scenario of tracking a vehicle traveling on a city street, correlation analysis might reveal that road width, traffic flow, and vehicle speed are highly sensitive factors highly correlated with the target's trajectory. Changes in these factors can directly affect the vehicle's trajectory and speed, thus requiring close attention during subsequent tracking. Such correlation analysis allows for a more precise understanding of the key factors influencing target tracking, providing more accurate and effective information support for subsequent processing and prediction.

[0023] S30: Obtain the time series matrix of the motion-sensitive elements of the set.

[0024] After determining the set of motion-sensitive elements, it is necessary to obtain the time series matrix of these sensitive elements. This step aims to capture and record the fluctuation and change patterns of the sensitive elements on the actual time axis.

[0025] A sensitive element time series matrix, simply put, is a matrix with time as the axis, sensitive elements as rows (or columns), and element values ​​(or element states) as elements. It records the actual value or state of each sensitive element at different points in time, thus reflecting the dynamic changes of these elements over time. For example, in a scenario of tracking a person moving in an indoor space, sensitive elements might include room layout, lighting conditions, and the person's movement speed. By deploying appropriate sensors or monitoring equipment, data on these sensitive elements can be collected in real time and organized and recorded in chronological order.

[0026] Over time, each sensitive element will have a corresponding series of data points. These data points, arranged chronologically, constitute the time-series data of the sensitive element. Then, organizing this time-series data according to the sensitive elements forms a sensitive element time-series matrix. In this way, the fluctuations and patterns of change of each sensitive element over time can be intuitively observed, providing strong data support for subsequent analysis and prediction. At the same time, the construction of the sensitive element time-series matrix also provides a foundation for subsequent historical sample group retrieval and statistical analysis of time-series information in confidence constraint regions.

[0027] S40: By configuring channels in the confidence constraint region, retrieve the concentrated motion prediction trajectories of historical sample groups that satisfy the motion profile and the time series matrix of the sensitive elements, and statistically analyze the time series information of the confidence constraint region.

[0028] Specifically, after obtaining the motion profile, sensitive element set, and its time series matrix, the system uses a confidence constraint region configuration channel to search for historical sample groups that meet these conditions, and extracts concentrated motion prediction trajectories from them to statistically analyze the time series information of the confidence constraint region. The confidence constraint region is a spatial range dynamically set based on the current motion profile and sensitive element time series matrix. It represents the possible motion area of ​​the target in the current scene. This area is derived from past experience and data analysis and has a certain degree of confidence, meaning that the target's motion trajectory within this area is considered reliable. The configuration channel is a dedicated retrieval and matching mechanism that can quickly find similar sample groups in the historical sample library based on the input motion profile and sensitive element time series matrix. These historical sample groups contain motion trajectories and sensitive element data recorded when tracking targets in various scenarios.

[0029] By traversing these historical sample groups, samples that highly match the current motion profile and the temporal matrix of sensitive elements can be identified, and their predicted motion trajectories can be extracted. These predicted motion trajectories represent the possible motion paths of targets similar to the current scene and target type at different historical moments. Subsequently, these predicted motion trajectories are statistically analyzed, with particular attention paid to their distribution and variation patterns within the confidence constraint region. Through calculation and analysis, the temporal information of the confidence constraint region can be derived, including the target's possible position, velocity, acceleration, and other motion parameters within that region, as well as the variation patterns of these parameters over time. For example, when tracking a vehicle traveling on a city street, configuring channels within the confidence constraint region may retrieve historical motion trajectories of similar vehicles that have previously traveled on that street. The portions of these trajectories within the confidence constraint region can then be extracted, and the temporal information such as the vehicle's possible velocity range and acceleration changes within that region can be statistically analyzed. This information can provide strong support for subsequent target tracking and prediction.

[0030] S50: By using the motion trajectory prediction channel bound to the motion profile, the target motion trajectory monitoring time sequence information and the preset element feature value matrix are processed to generate a motion prediction trajectory.

[0031] Preferably, during target tracking, a motion trajectory prediction channel bound to a motion profile can process the target motion trajectory monitoring time-series information and a preset feature value matrix to generate a predicted motion trajectory. This process combines real-time motion monitoring data with pre-set control parameters to achieve accurate prediction of the target's future motion trajectory.

[0032] Motion profiles, serving as the initial input for target tracking, contain basic target information and scene type, providing a foundation for subsequent predictions. The motion trajectory prediction channel, on the other hand, is a specially designed processing mechanism that combines real-time target motion trajectory monitoring information with a pre-defined feature value matrix. This pre-defined feature value matrix is ​​a set of feature values ​​containing all control elements, which may include the target's velocity, acceleration, direction, and position, as well as influencing factors in the scene such as lighting, occlusion, and noise. These feature values ​​are derived in advance through extensive experiments and data analysis, representing the general laws and characteristics of target motion in different scenes.

[0033] When the target's motion trajectory monitoring time-series information enters the motion trajectory prediction channel, the system compares and analyzes this real-time data with a preset feature value matrix. By calculating the similarity and difference between the real-time data and the preset feature values, the system can infer the target's motion state and trend in the current scene. For example, when tracking a vehicle driving on a city street, the system monitors the vehicle's speed, acceleration, direction, and other motion parameters in real time and compares these parameters with relevant feature values ​​in the preset feature value matrix. If the real-time data and preset feature values ​​match highly, the vehicle's future motion trajectory can be predicted based on these feature values. Ultimately, the motion trajectory prediction channel generates a predicted motion trajectory, representing the target's possible movement path over a future period. This predicted trajectory considers not only the target's current motion state but also various influencing factors in the scene, thus possessing high accuracy and reliability.

[0034] S60: When the motion prediction trajectory satisfies the confidence constraint region time sequence information, the motion prediction trajectory is output, and target tracking pre-configuration is performed.

[0035] Specifically, during target tracking, when the generated motion prediction trajectory satisfies the temporal information of the confidence constraint region, this prediction trajectory is considered reliable and valid, and is then output to perform target tracking pre-configuration. This process ensures the accuracy and stability of tracking.

[0036] The predicted motion trajectory is derived from a comprehensive analysis of the target's historical motion data, current motion state, and a preset feature value matrix. It represents the target's possible movement path over a future period. The confidence constraint region temporal information, on the other hand, is dynamically set based on the target's motion profile and the sensitive element temporal matrix. It defines the target's possible movement range and temporal patterns in the current scenario. When the predicted motion trajectory matches the confidence constraint region temporal information, it means that the predicted trajectory not only conforms to the target's current motion state but also to its motion patterns and possible range of change in the current scenario. Therefore, this predicted trajectory is considered reliable and is output for target tracking pre-configuration. For example, when tracking the flight path of a drone in a complex environment, the system generates a predicted motion trajectory in real time and dynamically sets the confidence constraint region temporal information based on the drone's flight speed, direction, altitude, and other motion parameters, as well as environmental influencing factors such as wind speed, wind direction, and obstacles. When the predicted trajectory matches the confidence constraint region temporal information, the system immediately outputs this predicted trajectory and adjusts the drone's flight attitude and path accordingly to ensure stable and accurate target tracking. This process enables continuous and stable tracking of targets, maintaining high tracking accuracy and reliability even in complex and changing environments.

[0037] In a preferred embodiment, performing correlation analysis based on the scene type and the target type to obtain a set of motion-sensitive elements includes: configuring preset element sets for the scene type and the target type via a user terminal, wherein the preset element sets correspond one-to-one with preset element feature values; retrieving several preset element feature value sets and several target motion monitoring trajectories from the preset element sets; performing correlation analysis on the preset element sets based on the several target motion monitoring trajectories and in conjunction with the several preset element feature value sets to obtain a preset element correlation set; and extracting the set of motion-sensitive elements from the preset element sets whose preset element correlation is greater than or equal to a correlation threshold based on the preset element correlation set.

[0038] For example, the user terminal refers to the terminal device or interface that interacts with the target tracking system or application. Users can configure parameters such as scene type, target type, and preset element set through the user terminal. These parameters will serve as the basis for system analysis and processing. Specifically, users need to configure the preset element set according to the specific scene type and target type. These preset element sets are predefined and contain various elements that may affect the target's movement. Each element corresponds one-to-one with a preset element feature value, which represents the quantitative performance of the element under specific conditions.

[0039] Subsequently, the system retrieves several preset feature value sets from the preset feature set. These sets represent combinations of feature values ​​for different elements under different conditions. Simultaneously, it collects several target motion monitoring trajectories, which are actual movement records of the targets over a past period. Based on this data foundation, a correlation analysis is performed on the preset feature set using these target motion monitoring trajectories and the preset feature value sets. The correlation analysis aims to identify which elements have a high correlation with the target motion trajectory, i.e., which element changes have a significant impact on the target motion trajectory. The analysis results form a preset element correlation set, which includes a correlation score between each element and the target motion trajectory. Finally, based on this preset element correlation set, the system extracts elements from the preset feature set whose correlation is greater than or equal to a correlation threshold to form a motion-sensitive element set. These elements are highly correlated with the target motion trajectory, and their changes will directly affect the target's motion trajectory.

[0040] For example, in a city street scenario, a user might configure a preset set of elements, including road width, traffic flow, and pedestrian density. Through retrieval and analysis, it's found that traffic flow and pedestrian density have a high correlation with the target's trajectory, so they are extracted as motion-sensitive elements. In this way, during subsequent target tracking, the system can focus on changes in these sensitive elements to more accurately predict the target's trajectory.

[0041] In a preferred embodiment, based on the several target motion monitoring trajectories and combined with the several preset element feature value sets, a correlation analysis is performed on the preset element set to obtain a preset element correlation set. This includes: calculating pairwise distances for the several target motion monitoring trajectories to obtain multiple dynamic time-normalized distances, performing dimensionless processing to obtain a baseline data sequence; calculating pairwise same-attribute deviations for the several preset element feature value sets to obtain multiple sets of preset element feature value deviation moduli, performing dimensionless processing on each set of preset element feature value deviation moduli to obtain multiple comparison data sequences; constructing a grey correlation matrix based on the baseline data sequence and the multiple comparison data sequences, and performing correlation analysis on the preset element set to obtain the preset element correlation set.

[0042] Optionally, the distances between each pair of the collected target motion monitoring trajectories are first calculated using dynamic time-warped distance, which effectively measures the similarity between different trajectories. After calculation, these distances are dimensionless to eliminate the influence of different dimensions on subsequent analysis, thus obtaining a baseline data sequence. Next, pairwise same-attribute deviation calculations are performed on several preset feature value sets. Same-attribute deviation calculation aims to identify the deviations of different feature values ​​on the same attribute, quantifying these differences by calculating the deviation modulus. Similarly, the deviation modulus of each set of preset feature values ​​is dimensionless to ensure data consistency and comparability, thereby obtaining multiple comparison data sequences.

[0043] A grey relational degree matrix is ​​constructed using a baseline data sequence and multiple comparative data sequences. Grey relational analysis is an effective data processing method that reveals the inherent relationships between data by calculating the correlation degree between various factors. In this matrix, each element represents the correlation degree between a feature in the preset feature set and the target trajectory. Finally, a comprehensive correlation analysis is performed on the preset feature set based on this grey relational degree matrix, resulting in a preset feature correlation degree set. The preset feature correlation degree set contains a correlation score between each preset feature and the target trajectory; the higher the score, the greater the influence of the feature on the target trajectory.

[0044] For example, in an intelligent transportation system, to analyze the impact of factors such as traffic flow and vehicle speed on vehicle trajectories, vehicle trajectory data from different time periods can be collected, and the dynamic time-normalized distances between them can be calculated. Simultaneously, eigenvalue data of factors such as traffic flow and vehicle speed can be collected, and the deviation modulus between them can be calculated. By constructing a grey relational matrix, it can be clearly seen which factors have the most significant impact on vehicle trajectories, thus providing strong support for traffic management and planning.

[0045] In a preferred embodiment, by configuring a confidence constraint region channel, a set of concentrated motion prediction trajectories of historical sample groups that satisfy the motion profile and the sensitive element time series matrix are retrieved, and the time series information of the confidence constraint region is statistically analyzed. This includes: retrieving historical sample groups that satisfy the motion profile; extracting the sample sensitive element time series record matrix set and the motion record trajectory set of the historical sample groups; traversing the sample sensitive element time series record matrix set and comparing it with the sensitive element time series matrix to obtain a sample consistency time zone set; traversing the sample consistency time zone set and advancing a preset time step to obtain a sample prediction time zone set; extracting a set of motion prediction trajectories from the set of motion record trajectories based on the sample prediction time zone set; extracting the concentrated motion prediction trajectories from the set of motion prediction trajectories, and statistically analyzing the time series information of the confidence constraint region.

[0046] In detail, historical sample groups matching the current motion profile are retrieved. These sample groups contain data recorded during target tracking in different scenarios in the past. From these historical sample groups, a set of time-series records of sensitive elements and a set of motion trajectory records are extracted. The set of time-series records of sensitive elements contains the sensitive element values ​​of each sample at different time points, while the set of motion trajectory records records the motion trajectory of each sample. These sets of time-series records of sensitive elements are traversed and compared with the current sensitive element time-series matrix. This comparison identifies samples that are highly consistent with the current sensitive element time-series matrix, forming a set of sample consistency time zones. These time zones represent periods in which the changes in the sensitive elements of the samples are highly similar to the current situation. This set of sample consistency time zones is traversed, and a preset time step is advanced to obtain a set of sample prediction time zones. These prediction time zones predict the possible future behavior patterns of the target based on the changes in the sensitive elements of historical samples. With these sets of sample prediction time zones, corresponding sets of motion prediction trajectories can be extracted from the set of motion trajectory records. These motion prediction trajectories represent the possible movement paths of the target in different prediction time zones. Finally, the concentrated motion prediction trajectories from these sets of motion prediction trajectories are extracted, and the temporal information of the confidence constraint area is statistically analyzed based on this. This information includes the target's possible position, velocity, acceleration, and other motion parameters within the confidence constraint area, as well as the variation patterns of these parameters over time. For example, in an urban traffic monitoring system, historical sample groups of vehicle travel under similar traffic flow and weather conditions are retrieved. By comparing the temporal record matrices of the sensitive elements of these samples with the current temporal record matrices of the sensitive elements, the system can identify samples that are highly consistent with the current situation and predict the possible future vehicle trajectories. Then, the temporal information of the confidence constraint area can be statistically analyzed based on these predicted trajectories, providing strong support for traffic management and planning.

[0047] In a preferred embodiment, traversing the set of sample consistency time zones and advancing a preset time step to obtain a set of sample prediction time zones includes: Step 1: Extracting a first sample consistency time zone from the set of sample consistency time zones; Step 2: If the first sample consistency time zone is empty, returning to Step 1; Step 3: If the first sample consistency time zone is not empty, advancing a preset time step to obtain a first sample prediction time zone, adding it to the set of sample prediction time zones, and then determining whether the traversal of the set of sample consistency time zones is complete. If complete, outputting the set of sample prediction time zones; otherwise, returning to Step 1.

[0048] Specifically, in the process of traversing the set of sample consistent time zones to obtain the set of sample prediction time zones, the first sample consistent time zone is extracted from the set of sample consistent time zones, which is the starting point for subsequent analysis. If the extracted first sample consistent time zone is empty, it means that no historical time zone matching the given sensitive element time series matrix has been found under the current conditions. At this time, the process returns to step one and continues to try to extract the next sample consistent time zone.

[0049] If the first consistent time zone is not empty, the next step is performed: advancing the preset time step. This step is crucial for predicting future behavior based on historical data; by setting a reasonable time step, the possible states of the target over a future period can be simulated. After advancing the time step, the first sample prediction time zone is obtained. This time zone represents the future target behavior pattern predicted based on the current consistent time zone. Subsequently, this first sample prediction time zone is added to the sample prediction time zone set, which is an important step in constructing the prediction time zone set. After adding, it is determined whether the sample consistent time zone set has been completely traversed. If the traversal is complete, it means that all possible consistent time zones have been processed and converted into prediction time zones. At this point, the entire sample prediction time zone set is output for subsequent analysis.

[0050] If the set of consistent time zones for the samples has not been fully traversed, the system will return to step one, continue to extract the next consistent time zone for the samples, and repeat the above process until all consistent time zones have been processed.

[0051] For example, in a weather forecasting system, the system might predict weather changes over a future period based on the time-series matrix of sensitive elements such as temperature and humidity from historical meteorological data. By traversing a set of consistent time zones, it can identify historical time zones similar to the current weather conditions and predict possible future weather patterns accordingly. Each time a consistent time zone is found, the system advances by a preset time step (such as one day, one week, etc.) to obtain the corresponding forecast time zone, and then combines these forecast time zones to form a complete weather forecast result.

[0052] In a preferred embodiment, extracting the concentrated motion prediction trajectories from the set of motion prediction trajectories and statistically analyzing the temporal information of the confidence constraint region includes: calculating pairwise trajectory similarity for the set of motion prediction trajectories to obtain multiple trajectory similarities; performing outlier trajectory deletion on the set of motion prediction trajectories based on the multiple trajectory similarities to obtain the concentrated motion prediction trajectories; statistically analyzing the set of motion positions of the concentrated motion prediction trajectories at the first moment, statistically analyzing the eight-directional boundary positions at the first moment, and constructing a polyhedral region at the first moment; and constructing the temporal information of the confidence constraint region based on the polyhedral region at the first moment up to the polyhedral region at the Nth moment.

[0053] Furthermore, in the process of extracting the concentrated motion prediction trajectories from the motion prediction trajectory set and statistically analyzing the temporal information of the confidence constraint region, the system first performs pairwise trajectory similarity calculation on the motion prediction trajectory set. The pairwise trajectory similarity calculation aims to measure the degree of similarity between different trajectories. By calculating multiple trajectory similarity values, these similarity values ​​reflect the similarity or difference between the trajectories and form the basis for subsequent processing.

[0054] Furthermore, outlier trajectory removal is performed on the motion prediction trajectory set based on trajectory similarity values. Outlier trajectories are those that differ significantly from other trajectories, which may be due to noise, anomalies, or erroneous data. By removing these outlier trajectories, more accurate and reliable centralized motion prediction trajectories can be obtained.

[0055] After obtaining the predicted trajectories of concentrated motion, the set of motion positions of these trajectories at different times is statistically analyzed. Specifically, the set of motion positions at the first moment is statistically analyzed, and the eight-directional boundary positions at that moment are further statistically analyzed. The eight-directional boundary positions refer to the farthest positions of the target in the eight main directions (such as east, south, west, north, southeast, northeast, southwest, and northwest), which together constitute a polyhedral region, representing the spatial range in which the target may appear at the first moment.

[0056] Subsequently, following this method, the set of motion positions and eight-directional boundary positions of the concentrated motion prediction trajectory at time 2, time 3, ... up to time N will be statistically analyzed, and corresponding polyhedral regions will be constructed. These polyhedral regions change over time, reflecting the spatial range in which the target may appear at different times. Finally, based on these polyhedral regions from time 1 to time N, temporal information of the confidence constraint region will be constructed. This information describes the spatial range in which the target may appear over a period of time and its changing patterns, providing important reference for subsequent tracking and prediction.

[0057] For example, in a drone tracking system, the system might calculate the similarity between multiple predicted trajectories and remove outliers that significantly deviate from the others. Then, it would analyze the position information of the remaining trajectories at different times to construct polyhedral regions where the target might appear at different times. The changing patterns of these regions can help the system predict the drone's future flight path and adjust its tracking strategy accordingly.

[0058] In a preferred embodiment, a motion trajectory prediction channel bound to the motion profile processes the target motion trajectory monitoring time-series information and a preset element feature value matrix to generate a motion prediction trajectory. This includes: collecting a dataset of preset element feature value records that satisfy the motion profile and a set of target motion trajectory record time-series information; dividing the target motion trajectory record time-series information set into two parts to obtain a front-end target motion trajectory record time-series information set and a back-end target motion trajectory record time-series information set, wherein any one of the back-end target motion trajectory record time-series information sets satisfies a preset time step; using the back-end target motion trajectory record time-series information set as supervision, and using the preset element feature value record dataset and the front-end target motion trajectory record time-series information set, training the motion trajectory prediction channel through machine learning, and binding it to the motion profile.

[0059] Specifically, in the process of generating the motion prediction trajectory, a dataset of preset feature values ​​that meet a specific motion profile and a set of time-series information of the target motion trajectory are first collected. The dataset of preset feature values ​​contains the quantified values ​​of various factors affecting the target motion (such as velocity, acceleration, direction, etc.) under different conditions, while the set of time-series information of the target motion trajectory records the actual motion path and time information of the target over a period of time.

[0060] The target motion trajectory recording time series information set is divided into a first-segment target motion trajectory recording time series information set and a second-segment target motion trajectory recording time series information set. The purpose of this division is to use the first-segment trajectory information as input and the second-segment trajectory information as supervision to train the motion trajectory prediction channel. Specifically, each record in the second-segment target motion trajectory recording time series information set satisfies a preset time step, which ensures the effectiveness and consistency of the supervision information.

[0061] During training, a preset set of feature value records and a set of time-series information on the preceding target motion trajectory are used as input, while a set of time-series information on the following target motion trajectory is used as supervision. A machine learning algorithm is then used to train the motion trajectory prediction channel. This channel learns and captures the complex relationship between the target motion trajectory and the preset feature values, thereby enabling the prediction of future motion trajectories. Finally, the trained motion trajectory prediction channel is bound to a specific motion profile. Thus, when new target motion trajectory monitoring time-series information and a preset feature value matrix are received, the corresponding predicted motion trajectory can be quickly generated through this channel.

[0062] For example, in an intelligent transportation system, the system might collect characteristic values ​​of vehicle speed, acceleration, and other features under different road and weather conditions, as well as the vehicle's actual driving trajectory information. This trajectory information is then divided into two segments, and the preceding and following segments are used to train a vehicle trajectory prediction channel. After training, this channel can quickly predict the future driving trajectory of vehicles based on new road conditions, weather conditions, and vehicle driving status, providing strong support for traffic management and planning.

[0063] The target tracking method combining multi-channel fusion provided by this invention has at least the following technical effects:

[0064] 1. By combining a multi-channel fusion target tracking method, accurate prediction of target motion trajectories is achieved. A set of motion-sensitive elements is obtained through correlation analysis, and a temporal matrix of these elements is constructed. Then, the concentrated motion prediction trajectories of historical sample groups are retrieved using confidence constraint region configuration channels, and the temporal information of the confidence constraint region is statistically analyzed. This process not only considers the inherent laws of target motion but also incorporates external environmental factors, such as scene type and target type, thereby significantly improving the accuracy and reliability of motion prediction. When the motion prediction trajectory satisfies the temporal information of the confidence constraint region, the predicted trajectory is output and target tracking pre-configuration is executed, ensuring the continuity and stability of tracking.

[0065] 2. In the process of extracting concentrated motion prediction trajectories, a pairwise trajectory similarity calculation method was adopted to effectively measure the degree of similarity between different trajectories. By setting a reasonable similarity threshold, outlier trajectories can be accurately identified and deleted, thereby retaining more representative concentrated motion prediction trajectories. This not only improves the purity of trajectory data but also provides a more reliable foundation for subsequent statistical analysis of time-series information in confidence constraint regions.

[0066] 3. By using a motion trajectory prediction channel bound to a motion profile, adaptive processing of the target motion trajectory monitoring time-series information and the preset feature value matrix is ​​achieved. This is accomplished by collecting a dataset of preset feature value records that satisfy the motion profile and a set of target motion trajectory record time-series information. The target motion trajectory record time-series information set is then bisected, and the latter part of the trajectory information is used as supervision to train the motion trajectory prediction channel. This process enables the prediction channel to adaptively learn and capture the complex relationship between the target motion trajectory and the preset feature values, thereby generating more accurate motion prediction trajectories. This adaptive training method not only improves the generalization ability of the prediction channel but also reduces the reliance on manual intervention, making the entire target tracking system more intelligent and automated.

[0067] Example 2:

[0068] like Figure 2 As shown, based on the same inventive concept as the target tracking method combining multi-channel fusion provided in Embodiment 1, this embodiment of the invention also provides a target tracking system combining multi-channel fusion, the system comprising:

[0069] The image acquisition module 11 is used to obtain a motion image of the tracked target, wherein the motion image includes scene type and target type.

[0070] The correlation analysis module 12 is used to perform correlation analysis based on the scene type and the target type to obtain a set of motion-sensitive elements.

[0071] The time series analysis module 13 is used to obtain the time series matrix of the sensitive elements of the motion sensitive element set.

[0072] The channel configuration module 14 is used to configure channels through confidence constraint regions, retrieve concentrated motion prediction trajectories of historical sample groups that satisfy the motion profile and the time series matrix of sensitive elements, and statistically analyze the time series information of the confidence constraint regions.

[0073] The trajectory prediction module 15 is used to process the target motion trajectory monitoring time sequence information and the preset element feature value matrix through the motion trajectory prediction channel bound to the motion profile, and generate a motion prediction trajectory.

[0074] The tracking configuration module 16 is used to output the motion prediction trajectory and perform target tracking pre-configuration when the motion prediction trajectory meets the confidence constraint region time sequence information.

[0075] Furthermore, the correlation analysis module 12 is also used to perform the following steps:

[0076] Through the user terminal, a preset element set for the scene type and the target type is configured, wherein the preset element set corresponds one-to-one with the preset element feature value; several preset element feature value sets and several target motion monitoring trajectories are retrieved from the preset element set; based on the several target motion monitoring trajectories and combined with the several preset element feature value sets, a correlation analysis is performed on the preset element set to obtain a preset element correlation set; based on the preset element correlation set, the motion-sensitive element set with a preset element correlation degree greater than or equal to the correlation degree threshold is extracted from the preset element set.

[0077] Furthermore, the correlation analysis module 12 is also used to perform the following steps:

[0078] The distances between each pair of the motion monitoring trajectories of the aforementioned targets are calculated to obtain multiple dynamic time-normalized distances, which are then dimensionlessly processed to obtain a baseline data sequence. The pairwise same-attribute deviations of the feature value sets of the aforementioned preset elements are calculated to obtain multiple sets of preset element feature value deviation moduli. Each set of preset element feature value deviation moduli is then dimensionlessly processed to obtain multiple comparison data sequences. Based on the baseline data sequence and the multiple comparison data sequences, a grey relational degree matrix is ​​constructed, and the relational degree of the preset element set is analyzed to obtain the preset element relational degree set.

[0079] Furthermore, the channel configuration module 14 is also used to perform the following steps:

[0080] Retrieve historical sample groups that satisfy the motion profile; extract the sample sensitive element time-series record matrix set and motion record trajectory set of the historical sample groups; traverse the sample sensitive element time-series record matrix set and compare it with the sensitive element time-series matrix to obtain a sample consistency time zone set; traverse the sample consistency time zone set and advance a preset time step to obtain a sample prediction time zone set; extract a motion prediction trajectory set from the motion record trajectory set based on the sample prediction time zone set; extract the concentrated motion prediction trajectory from the motion prediction trajectory set and statistically analyze the confidence constraint region time-series information.

[0081] Furthermore, the channel configuration module 14 is also used to perform the following steps:

[0082] Step 1: Extract the first sample consistency time zone from the sample consistency time zone set; Step 2: If the first sample consistency time zone is empty, return to Step 1; Step 3: If the first sample consistency time zone is not empty, advance by a preset time step to obtain the first sample prediction time zone, add it to the sample prediction time zone set, and determine whether the sample consistency time zone set has been traversed completely. If it has, output the sample prediction time zone set; otherwise, return to Step 1.

[0083] Furthermore, the channel configuration module 14 is also used to perform the following steps:

[0084] The set of predicted motion trajectories is subjected to pairwise trajectory similarity calculation to obtain multiple trajectory similarities. Based on the multiple trajectory similarities, outlier trajectory deletion is performed on the set of predicted motion trajectories to obtain the set of concentrated predicted motion trajectories. The set of motion positions at the first moment of the concentrated predicted motion trajectories is statistically analyzed, and the eight-directional boundary positions at the first moment are statistically analyzed to construct the polyhedral region at the first moment. This process continues until the set of motion positions at the Nth moment of the concentrated predicted motion trajectories is statistically analyzed, and the eight-directional boundary positions at the Nth moment are statistically analyzed to construct the polyhedral region at the Nth moment. The temporal information of the confidence constraint region is constructed based on the polyhedral region at the first moment up to the polyhedral region at the Nth moment.

[0085] Furthermore, the trajectory prediction module 15 is also used to perform the following steps:

[0086] Collect a dataset of preset feature value records that satisfy the motion profile and a set of time-series information of target motion trajectory records; divide the set of time-series information of target motion trajectory records into two parts to obtain a set of time-series information of the preceding target motion trajectory records and a set of time-series information of the following target motion trajectory records, wherein any one of the time-series information of the following target motion trajectory records satisfies a preset time step; using the set of time-series information of the following target motion trajectory records as supervision, and using the preset feature value record dataset and the set of time-series information of the preceding target motion trajectory records, train the motion trajectory prediction channel through machine learning and bind it to the motion profile.

[0087] Through the foregoing detailed description of a target tracking method combining multi-channel fusion, those skilled in the art can clearly understand the target tracking system combining multi-channel fusion in this embodiment. As the system disclosed in the embodiment corresponds to the method disclosed in the embodiment, the description is relatively simple, and relevant parts can be referred to the method section.

[0088] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A target tracking method combining multi-channel fusion, characterized in that, include: Obtain a motion profile of the tracked target, wherein the motion profile includes scene type and target type; Perform correlation analysis based on the scene type and the target type to obtain a set of motion-sensitive elements; Obtain the time-series matrix of the motion-sensitive elements in the set of motion-sensitive elements; By configuring channels in the confidence constraint region, we retrieve concentrated motion prediction trajectories of historical sample groups that satisfy the motion profile and the time series matrix of the sensitive elements, and statistically analyze the time series information of the confidence constraint region. By using the motion trajectory prediction channel bound to the motion profile, the target motion trajectory monitoring time series information and the preset element feature value matrix are processed to generate a motion prediction trajectory. When the motion prediction trajectory satisfies the time sequence information of the confidence constraint region, the motion prediction trajectory is output, and target tracking pre-configuration is performed; Specifically, a correlation analysis is performed based on the scene type and the target type to obtain a set of motion-sensitive elements, including: Through the user terminal, a preset element set for the scene type and the target type is configured, wherein the preset element set corresponds one-to-one with the preset element feature value; Retrieve a set of preset feature values ​​of the preset feature set and a set of target motion monitoring trajectories; Based on the motion monitoring trajectories of the aforementioned targets, and combined with the feature value set of the aforementioned preset elements, a correlation analysis is performed on the preset element set to obtain a preset element correlation set. Based on the preset element correlation set, extract the motion-sensitive element set from the preset element set whose preset element correlation is greater than or equal to the correlation threshold; Specifically, by configuring channels within the confidence constraint region, concentrated motion prediction trajectories of historical sample groups that satisfy the motion profile and the time-series matrix of sensitive elements are retrieved, and the time-series information of the confidence constraint region is statistically analyzed, including: Retrieve historical sample groups that satisfy the motion profile; Extract the time-series record matrix set and motion record trajectory set of the sample sensitive elements of the historical sample group; Traverse the set of time series record matrices of the sensitive elements of the sample and compare them with the time series matrix of the sensitive elements to obtain the set of time zones with sample consistency; Traverse the set of time zones with consistent sample data, advance by a preset time step, and obtain the set of predicted time zones for the sample data. Based on the sample predicted time zone set, extract the motion prediction trajectory set from the motion record trajectory set; Extract the concentrated motion prediction trajectory from the set of motion prediction trajectories, and statistically analyze the temporal information of the confidence constraint region.

2. The method as described in claim 1, characterized in that, Based on the aforementioned target motion monitoring trajectories, and combined with the aforementioned preset element feature value sets, a correlation analysis is performed on the preset element set to obtain a preset element correlation set, including: The pairwise distances of the motion monitoring trajectories of the aforementioned targets are calculated to obtain multiple dynamic time-normalized distances, which are then dimensionlessly processed to obtain a baseline data sequence. The pairwise same attribute deviation is calculated for the set of preset feature values ​​to obtain multiple sets of preset feature value deviation modulus. The dimensionless processing is performed on each set of preset feature value deviation modulus to obtain multiple comparison data sequences. Based on the baseline data sequence and the multiple comparison data sequences, a grey relational degree matrix is ​​constructed, and the relational degree analysis is performed on the preset element set to obtain the preset element relational degree set.

3. The method as described in claim 1, characterized in that, Traverse the set of sample consistency time zones, advance by a preset time step, and obtain the set of sample prediction time zones, including: Step 1: Extract the first sample consistency time zone from the set of sample consistency time zones; Step 2: If the consistency time zone of the first sample is empty, return to Step 1; Step 3: When the first sample consistency time zone is not empty, advance the preset time step to obtain the first sample prediction time zone, add it to the sample prediction time zone set, and determine whether the sample consistency time zone set has been traversed. If it has, output the sample prediction time zone set; otherwise, return to Step 1.

4. The method as described in claim 1, characterized in that, Extract the set of concentrated motion prediction trajectories from the set of motion prediction trajectories, and statistically analyze the temporal information of the confidence constraint region, including: Perform pairwise trajectory similarity calculations on the set of motion prediction trajectories to obtain multiple trajectory similarities; Based on the multiple trajectory similarities, outlier trajectory deletion is performed on the set of motion prediction trajectories to obtain the concentrated motion prediction trajectories; The set of motion positions at the first moment of the predicted concentrated motion trajectory is statistically analyzed, the eight-directional boundary positions at the first moment are statistically analyzed, and a polyhedral region at the first moment is constructed. Until the Nth time of the statistical analysis of the predicted motion trajectory, the set of motion positions at the Nth time is statistically analyzed, the eight-directional boundary positions at the Nth time are statistically analyzed, and the polyhedral region at the Nth time is constructed. The confidence constraint region temporal information is constructed based on the polyhedral region from the first time step up to the polyhedral region at the Nth time step.

5. The method as described in claim 1, characterized in that, By processing the target motion trajectory monitoring time-series information and the preset element feature value matrix through the motion trajectory prediction channel bound to the motion profile, a motion prediction trajectory is generated, including: Collect a dataset of preset feature values ​​that satisfy the motion profile and a set of time-series information of the target motion trajectory; The target motion trajectory recording time sequence information set is divided into two parts to obtain a front-end target motion trajectory recording time sequence information set and a back-end target motion trajectory recording time sequence information set, wherein any one of the back-end target motion trajectory recording time sequence information sets satisfies a preset time step. Using the set of time-series information of the target motion trajectory records in the later stage as supervision, and using the preset feature value record dataset and the set of time-series information of the target motion trajectory records in the earlier stage, the motion trajectory prediction channel is trained through machine learning and bound to the motion profile.

6. A target tracking system combining multi-channel fusion, characterized in that, The system for implementing the target tracking method combining multi-channel fusion as described in any one of claims 1-5, the system comprising: The image acquisition module is used to obtain a motion image of the tracked target, wherein the motion image includes scene type and target type; The correlation analysis module is used to perform correlation analysis based on the scene type and the target type to obtain a set of motion-sensitive elements; The time-series analysis module is used to obtain the time-series matrix of the sensitive elements in the motion-sensitive element set; The channel configuration module is used to configure channels through confidence constraint regions, retrieve concentrated motion prediction trajectories of historical sample groups that satisfy the motion profile and the time series matrix of sensitive elements, and statistically analyze the time series information of the confidence constraint regions. The trajectory prediction module is used to process the target motion trajectory monitoring time series information and the preset element feature value matrix through the motion trajectory prediction channel bound to the motion profile, and generate a motion prediction trajectory. The tracking configuration module is used to output the motion prediction trajectory and perform target tracking pre-configuration when the motion prediction trajectory meets the time series information of the confidence constraint region.

Images