Adaptive context-aware vehicle target tracking and recognition system

Abstract

The present invention relates to the technical field of vehicle target tracking and recognition, and discloses an adaptive context-aware vehicle target tracking and recognition system. The system comprises a multimodal information acquisition module, a target feature extraction module, a scene dynamics perception module, a context association analysis module, and a tracking strategy generation module. The multimodal information acquisition module performs multi-dimensional data acquisition and triggers related modules; the target feature extraction module extracts multiple vehicle features and evaluates trackability; the scene dynamics perception module monitors system parameters and predicts dynamic perception effectiveness; the context association analysis module jointly analyzes the evaluation values ​​and generates control signals; and the tracking strategy generation module generates tracking maintenance and computing power allocation parameters based on the control signals. This system can adapt to complex dynamic scenarios and improve the stability and accuracy of vehicle target tracking and recognition.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle target tracking and recognition, and in particular to an adaptive context-aware vehicle target tracking and recognition system. Background Art

[0002] With the rapid development of intelligent transportation systems, vehicle target tracking and recognition technologies have been widely used in scenarios such as autonomous driving, traffic monitoring, and vehicle-road collaboration. However, current vehicle target tracking and recognition systems still face numerous challenges in complex and dynamic environments. In practical applications, traffic scenarios are often highly dynamic and uncertain. For example, on urban roads during rush hour, vehicles are densely populated and their driving states are highly variable, with frequent acceleration, deceleration, lane changes, and overtaking. Furthermore, vehicles may be affected by weather conditions (such as rain, fog, and strong sunlight), causing significant changes in their visual characteristics. Traditional tracking systems often rely on a single visual sensor to collect data, making it difficult to fully capture the dynamic information of vehicles. When faced with occlusion, sudden changes in lighting conditions, and other conditions, tracking drift and even target loss are prone to occur.

[0003] Some existing systems have attempted to incorporate multi-sensor fusion technology, but these technologies have limitations in data processing and feature extraction. Most systems employ fixed feature extraction methods, targeting only a subset of vehicle features and failing to dynamically adjust their extraction strategies based on changing scenarios. For example, some systems focus solely on the vehicle's outline and color, ignoring the value of texture details in complex environments. This makes it difficult to accurately distinguish target vehicles in scenes with numerous similar vehicles, reducing tracking accuracy.

[0004] Existing systems lack flexibility in allocating computing resources when processing real-time data. With the increasing number of vehicles and improved data collection accuracy, the system's real-time data traffic has increased dramatically, and local computing load and communication latency have become increasingly prominent. Traditional systems typically use a fixed computing power allocation model, unable to dynamically adjust the allocation of edge and cloud computing resources based on real-time data traffic and computing load. This results in slower system response times, increased tracking latency, and even data processing congestion under high load conditions.

[0005] Existing systems underutilize contextual information, lacking in-depth awareness and contextual analysis of dynamic scene changes. For example, tracking performance can be significantly impacted when the system is exposed to varying communication environments or computing conditions, but existing systems often fail to detect these changes and adjust accordingly. Contextual information, including ambient lighting, road conditions, and communication signal strength, is closely tied to the vehicle's tracking status. Ignoring this information reduces the system's adaptability in complex scenarios, making it difficult to maintain stable tracking.

[0006] Existing tracking systems have a relatively simple strategy generation mechanism and lack multi-level optimization capabilities. During the tracking process, when anomalies such as target feature blur or occlusion occur, the system cannot quickly generate an effective response strategy and must rely on a pre-set fixed algorithm. This results in insufficient tracking stability and robustness. For example, on a highway, when a vehicle is traveling quickly and briefly obscured, existing systems may lose track of the target due to an inability to adjust tracking parameters in a timely manner, affecting subsequent tracking results.

[0007] The current vehicle target tracking and recognition system has obvious defects in dynamic scene adaptability, multimodal information fusion, dynamic allocation of computing power, and contextual association analysis, making it difficult to meet the needs of high-precision and high-stability tracking and recognition of vehicle targets in complex traffic environments. Summary of the Invention

[0008] The purpose of the present invention is to provide an adaptive context-aware vehicle target tracking and recognition system to solve the problems raised in the above background technology.

[0009] To achieve the above objectives, the present invention provides an adaptive context-aware vehicle target tracking and recognition system, the system comprising:

[0010] The multimodal information acquisition module is used to collect the vehicle's visual feature data, motion parameter data, and environmental context information in multiple dimensions to obtain a multi-source scene data set. Based on the multi-source scene data set, the module quantitatively identifies abnormal tracking states and generates tracking warning signals. The generated tracking warning signals trigger the scene dynamic perception module and the context association analysis module.

[0011] The target feature extraction module is used to extract the vehicle's contour feature parameters, color distribution features, and texture detail features, dynamically evaluate the target's trackability, and obtain a feature evaluation benchmark value;

[0012] The scene dynamic perception module is used to monitor the system's real-time data traffic, local computing load, and communication delay parameters, and to predict and analyze the scene's dynamic perception performance to obtain a dynamic perception evaluation value.

[0013] The context-related analysis module receives the feature evaluation baseline value and the dynamic perception evaluation value, performs a joint analysis on the system's context-related effectiveness, and generates edge control signals and cloud-side reinforcement signals;

[0014] The tracking strategy generation module is used to receive edge control signals and cloud reinforcement signals, perform multi-level optimization strategy analysis, and generate vehicle tracking maintenance parameters and computing power allocation parameters.

[0015] Preferably, the quantitative identification of abnormal tracking status based on the scenario multi-source data set includes:

[0016] By extracting the target ambiguity, color offset and contour integrity from the vehicle's visual feature data, the ambiguity value, color offset value and contour integrity value are obtained. The values ​​of the three are extracted and weighted to calculate the contribution of tracking anomaly.

[0017] By extracting the signal interruption frequency, parameter matching degree and data packet loss rate from the system's sensor communication status parameters, the communication interruption value, parameter matching value and data packet loss value are obtained, and they are marked as tracking stability characteristic values. A tracking stability threshold is set, and the characteristic value is compared and analyzed with the threshold. When the characteristic value is less than the threshold, the sensor is marked as an abnormal sensor. The ratio of the number of abnormal sensors to the total number of sensors in the current system is counted to obtain the communication anomaly rate. At the same time, the light intensity change and the frequency of occlusion in the environmental context information are extracted, and dynamic simulation calculations are performed to obtain the environmental interference assessment value.

[0018] The values ​​of tracking anomaly contribution, communication anomaly rate and environmental interference assessment value are multiplied by the corresponding weight coefficients and added together to obtain the abnormal state fusion value, which is compared with the preset abnormality threshold. If the fusion value is higher than the threshold, a tracking warning signal is generated.

[0019] Preferably, the dynamic evaluation of the traceability of the target includes:

[0020] By analyzing the spatial distribution of the vehicle's contour feature parameters, the corresponding contour change distribution map and occlusion risk assessment map are generated;

[0021] Extract the reference distribution map of the standard tracking vehicle profile from the system database, perform a topological comparison between the target vehicle's profile change distribution map and the reference distribution map, calculate the profile matching degree of the two, and perform normalization processing to obtain the target tracking potential index;

[0022] Extract the occlusion occurrence frequency and feature loss ratio from the vehicle occlusion risk assessment map and mark them as occlusion feature values ​​respectively;

[0023] Extract the standard occlusion risk threshold from the system database, calculate the difference between the occlusion feature value and the threshold, and obtain the traceability deviation;

[0024] The target tracking potential index and the trackability deviation value are weighted and fused to obtain the feature evaluation benchmark value.

[0025] Preferably, the predictive analysis of the dynamic perception efficiency of the scene includes:

[0026] By extracting the number of image frames, motion parameters and log storage capacity from the real-time data flow of the system, a data flow parameter set is obtained;

[0027] Extract historical tracking data of similar scenarios from the system database, build an efficiency prediction model based on a dynamic balance algorithm, input the traffic parameter set into the model, and output the data processing rate, computing power utilization, and latency fluctuation rate within the target time interval;

[0028] The data processing rate, computing power utilization, and delay fluctuation rate are normalized and calculated to obtain the dynamic perception evaluation value.

[0029] Preferably, the joint analysis of the context-related effectiveness of the system includes:

[0030] Retrieve the system's abnormal tracking state quantification results, set its correction factor, and obtain the tracking impact correction value through calculation and processing;

[0031] Normalizing the feature evaluation benchmark value, the dynamic perception evaluation value, and the tracking impact correction value to obtain a contextual correlation evaluation value;

[0032] Set the context association evaluation threshold. If the association evaluation value is greater than or equal to the threshold, an edge control signal is generated; if it is less than the threshold, a cloud reinforcement signal is generated.

[0033] Preferably, the multi-level optimization strategy analysis includes:

[0034] If an edge control signal is captured, the tracking maintenance instruction is triggered, and the vehicle's tracking window size and feature matching accuracy are dynamically adjusted according to the instruction to generate tracking maintenance parameters;

[0035] If a cloud reinforcement signal is captured, the computing power allocation instruction will be triggered. Based on the instruction, the system's local storage capacity and cloud computing tasks will be dynamically planned to generate computing power allocation parameters.

[0036] Preferably, the system further comprises:

[0037] The permission control module is used to monitor the system's tracking operation permissions in real time, extract the operator's identification, permitted operation scope and illegal operation characteristics, and generate permission association evaluation values;

[0038] The context association analysis module further combines the authority association evaluation value to perform an operation authority constraint analysis on the context association effectiveness.

[0039] Preferably, the real-time monitoring of the tracking operation authority of the system includes:

[0040] By analyzing the license data of the system during operation control, the permission type and historical violation records of the operating subject are obtained;

[0041] Match the permission type with the preset operation permission list and calculate the permission type deviation;

[0042] Calculate the difference between the frequency of unauthorized operations and the total number of operations in historical violation records to obtain the abnormal operation behavior index;

[0043] The permission type deviation degree and the operation behavior abnormality index are weighted and fused to generate the permission association evaluation value.

[0044] Preferably, the system further comprises:

[0045] The dynamic adjustment module is used to adjust the resource configuration frequency of system operation and maintenance in real time based on the generated tracking maintenance parameters and computing power allocation parameters, and feed the adjusted parameters back to the multimodal information acquisition module to form a closed-loop optimization process.

[0046] Preferably, the adjustment process of the dynamic adjustment module includes:

[0047] If the tracking maintenance parameters involve adjusting the tracking window size, the sensor scheduling parameters in the resource configuration are adjusted synchronously;

[0048] If the computing power allocation parameters involve cloud computing task optimization, the network bandwidth allocation parameters in the resource configuration will be adjusted simultaneously;

[0049] Input the adjusted parameters into the multimodal information acquisition module and restart the optimization and verification process.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] The multimodal information acquisition module is capable of multi-dimensionally collecting the vehicle's visual feature data, motion parameter data, and environmental context information, overcoming the limitations of traditional systems that rely on a single sensor. This fusion of multi-source data allows the system to more comprehensively capture the vehicle's characteristic information in different scenarios. Even in situations with dense traffic, large lighting changes, or occlusion, the system can complement multi-dimensional data to reduce tracking errors caused by insufficient single data sources. At the same time, the module can quantitatively identify abnormal tracking states and trigger the corresponding modules, enabling timely response to abnormal conditions during the tracking process and avoiding the continued impact of abnormal conditions.

[0052] The target feature extraction module extracts the vehicle's contour feature parameters, color distribution characteristics, and texture detail features, and dynamically evaluates the target's trackability, breaking through the constraints of traditional fixed feature extraction methods. By comprehensively considering multiple feature parameters, the system can more accurately characterize the characteristics of the target vehicle, improving its ability to distinguish in scenes with a large number of similar vehicles. The dynamic evaluation mechanism enables the system to adjust the tracking strategy in real time based on changes in target features. When the target features become blurred or changed, it can promptly detect and provide reference for subsequent tracking strategy adjustments, enhancing the system's ability to adapt to dynamic changes in target features.

[0053] The scenario dynamic perception module monitors the system's real-time data traffic, local computing load, and communication latency parameters, and performs predictive analysis of the dynamic scenario perception performance, addressing the rigid computing power allocation issues of traditional systems. By monitoring key system operating parameters in real time, the system can proactively detect changing trends in data processing pressure and communication latency, providing a basis for subsequent computing power allocation and policy adjustments. This predictive analysis capability enables the system to prepare for surges in data traffic or excessive computing loads, avoiding tracking delays or data loss caused by insufficient resources and ensuring stable system operation in dynamic scenarios.

[0054] The contextual correlation analysis module receives feature evaluation baseline values ​​and dynamic perception evaluation values, performs a joint analysis of the system's contextual correlation effectiveness, and generates edge control signals and cloud reinforcement signals, achieving synergy and linkage between the system's modules. Traditional systems often isolate the work of each module and lack effective correlation analysis. However, this module, by comprehensively considering the evaluation results of target characteristics and scene dynamics, can more comprehensively grasp the overall operating status of the system. Based on the control signals generated, edge computing and cloud computing tasks can be rationally allocated, allowing the edge to quickly handle tasks with high real-time requirements, while the cloud handles complex data analysis and policy optimization, improving the overall operational efficiency of the system.

[0055] The tracking strategy generation module receives edge control signals and cloud reinforcement signals, performs multi-level optimization strategy analysis, and generates vehicle tracking maintenance parameters and computing power allocation parameters, overcoming the limitations of traditional system strategies that often rely on a single strategy. This multi-level optimization strategy enables the system to generate targeted tracking parameters and computing power allocation plans based on different scenarios and target states. For example, when target features are clear and the scene is stable, a relatively simple tracking strategy is adopted and less computing power is allocated. However, when target features are blurred or the scene is complex, more complex tracking maintenance parameters are activated and computing power support is increased, thereby improving the utilization efficiency of computing resources while ensuring tracking accuracy.

[0056] Through the collaborative work of various modules, the system can achieve accurate and stable tracking and identification of vehicle targets in complex dynamic environments, effectively addressing the shortcomings of traditional systems in multi-source data fusion, feature extraction adaptability, and computing power allocation flexibility, and improving the overall performance of vehicle target tracking and identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is a working principle diagram of the adaptive context-aware vehicle target tracking and recognition system of the present invention;

[0058] Figure 2 Flowchart for quantitative identification of abnormal tracking status;

[0059] Figure 3 Flowchart for dynamic evaluation of target traceability;

[0060] Figure 4 Flowchart for prediction of scene dynamic perception effectiveness;

[0061] Figure 5 Flowchart for the context-sensitive joint analysis. DETAILED DESCRIPTION

[0062] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0063] See also Figure 1-Figure 5 The present invention provides an adaptive context-aware vehicle target tracking and recognition system. The system includes a multimodal information acquisition module, a target feature extraction module, a scene dynamics perception module, a context association analysis module, and a tracking strategy generation module. The system is described in detail below with specific details:

[0064] The multimodal information acquisition module collects a vehicle's visual feature data, motion parameter data, and environmental context information from multiple dimensions to generate a multi-source scene data set. Based on this multi-source scene data set, it quantitatively identifies abnormal tracking states and generates tracking warning signals. These signals are then used to trigger the scene dynamic perception module and the context association analysis module.

[0065] The target feature extraction module is used to extract the vehicle's contour feature parameters, color distribution features and texture detail features, dynamically evaluate the target's trackability, and obtain feature evaluation benchmark values.

[0066] The scene dynamic perception module is used to monitor the system's real-time data traffic, local computing load and communication delay parameters, predict and analyze the scene's dynamic perception performance, and obtain a dynamic perception evaluation value.

[0067] The context association analysis module is used to receive feature evaluation benchmark values ​​and dynamic perception evaluation values, conduct a joint analysis of the system's context association effectiveness, and generate edge control signals and cloud reinforcement signals.

[0068] The tracking strategy generation module is used to receive edge control signals and cloud reinforcement signals, perform multi-level optimization strategy analysis, and generate vehicle tracking maintenance parameters and computing power allocation parameters.

[0069] Example 1: The process of quantitatively identifying abnormal tracking status by the multimodal information acquisition module begins with in-depth analysis of vehicle visual feature data.

[0070] First, the team focused on target blur, color offset, and outline integrity among the visual features, extracting these features through a specific image parsing algorithm. Target blur is converted into a quantifiable blur value by analyzing the sharpness of vehicle edges and pixel gradient variations in the image. Color offset is calculated by comparing the differences in the RGB color space distribution of the vehicle in different frames. Outline integrity is determined based on the continuity of the vehicle's outline and the proportion of missing parts in the image. After obtaining these three values, they are weighted according to a preset ratio to determine the tracking anomaly contribution. This value reflects the combined impact of the visual features that may cause tracking anomalies.

[0071] While obtaining the contribution to tracking anomalies, the system's sensor communication status parameters are simultaneously analyzed. Signal interruption frequency, parameter matching, and data packet loss rate are extracted from the sensor's real-time communication data. These parameters are converted into communication interruption values, parameter matching values, and data packet loss values, respectively. These three values ​​are then integrated and labeled as tracking stability characteristic values. A tracking stability threshold is pre-set within the system, based on the sensor's normal operating parameter range. The calculated tracking stability characteristic value is compared with this threshold. When the characteristic value is less than the threshold, it indicates that the corresponding sensor has an anomaly during communication, and the sensor is marked as an abnormal sensor. The number of sensors marked as abnormal in the current system is counted and its ratio to the total number of sensors in the system is calculated to obtain the communication anomaly rate, which intuitively reflects the abnormal conditions at the sensor communication level.

[0072] At the same time, changes in illumination intensity and the frequency of occlusions within the environmental context are collected and analyzed. Light intensity changes are captured in real time by a light sensor, which measures illumination values ​​at different points in time and calculates the amplitude of illumination fluctuations per unit time. The frequency of occlusions is determined by comparing consecutive frames of imagery and counting the number of times the vehicle is obscured by other objects. Dynamic simulations are performed combining these two data points, considering the impact of illumination changes on image acquisition quality and the interference of occlusions on target recognition continuity. This ultimately yields an environmental interference assessment value, which reflects the degree to which external environmental factors interfere with the tracking process.

[0073] After completing the calculation of the tracking anomaly contribution, communication anomaly rate, and environmental interference assessment value, these three values ​​are assigned corresponding weight coefficients. These coefficients are set according to the actual impact of each factor on the tracking anomaly in different scenarios. The three values ​​are multiplied by their respective weight coefficients and summed to obtain the abnormal state fusion value. The system has a preset abnormality threshold, which is determined based on the critical value of normal tracking and abnormal tracking in a large amount of historical tracking data. The abnormal state fusion value is compared with the abnormality threshold. If the fusion value is higher than the threshold, it indicates that the tracking state of the current system has become obviously abnormal. At this time, the multimodal information acquisition module generates a tracking warning signal, which will trigger the scene dynamic perception module and the context association analysis module to enter the corresponding working state to deal with possible tracking anomalies. The entire process achieves accurate quantitative identification of the system's abnormal tracking state through multi-dimensional data collection and comprehensive analysis, providing a reliable basis for subsequent tracking adjustments.

[0074] Example 2: When the target feature extraction module dynamically evaluates the trackability of the target, it conducts spatial distribution analysis on the vehicle's contour feature parameters. By performing edge detection and contour extraction on the collected vehicle images, the contour data of the vehicle at different angles and distances is obtained, including the coordinates of the key points of the contour, the direction of the lines and the overall shape. Based on these data, a corresponding contour change distribution map is generated, which can intuitively show the trend of the vehicle contour changing over time and space; at the same time, combined with the relative position relationship between the vehicle and the surrounding objects in the continuous frame images, the possible occlusion area and occlusion probability are analyzed to generate an occlusion risk assessment map, which can reflect the potential risk distribution of the vehicle being occluded during the tracking process.

[0075] A reference distribution map of vehicle contours for standard tracking is retrieved from the system database. This map is constructed based on a large amount of vehicle contour data under standard operating conditions, covering typical contour features of different vehicle models and postures. The target vehicle's contour change distribution map is topologically compared with the reference distribution map. The contour matching degree is calculated by calculating the number of matching key points, line similarity, and overall morphological consistency between the two. This contour matching degree is normalized, i.e., converted to a preset range of values, to form a target tracking potential index. This index reflects the degree of conformity between the target vehicle contour and the standard contour and serves as one of the basic indicators for evaluating target trackability.

[0076] The occlusion frequency and feature loss ratio are extracted from the vehicle's occlusion risk assessment map, and these two parameters are marked as occlusion feature values. The occlusion frequency refers to the number of times the vehicle is occluded per unit time, and the feature loss ratio refers to the proportion of key vehicle features (such as license plates and vehicle model identification) that are obscured when occlusion occurs. The standard occlusion risk threshold is extracted from the system database. This threshold is determined based on the maximum occlusion level in historical tracking data that does not affect normal tracking. The difference between the extracted occlusion feature value and the standard occlusion risk threshold is calculated to obtain the trackability deviation. This deviation reflects the difference between the actual occlusion situation and the standard threshold, and can reflect the degree of influence of occlusion factors on target trackability.

[0077] The target tracking potential index and trackability deviation values ​​are weighted and fused according to preset weights. The weighting ratio takes into account the actual effects of contour matching and occlusion in different tracking scenarios. This fusion calculation results in a feature evaluation baseline value that combines both contour characteristics and occlusion risk, serving as a comprehensive indicator for measuring target trackability.

[0078] When the scene dynamic perception module predicts and analyzes the dynamic perception performance of a scene, it first extracts the number of image frames, motion parameters, and log storage capacity from the system's real-time data flow. Image frames refer to the number of vehicle images collected per unit time, motion parameters refer to the number of parameters describing the vehicle's motion state (such as speed, acceleration, and azimuth), and log storage capacity refers to the storage size of the log data generated during system operation. These three parameters are integrated to form the data flow parameter set.

[0079] Extract historical tracking data for similar scenarios from the system database. This data includes information such as real-time data traffic, processing efficiency, and system load in similar scenarios in the past. Build an efficiency prediction model based on a dynamic balance algorithm. This model establishes a prediction function by learning the correlation between data traffic and processing efficiency in historical data. Input the current set of traffic parameters into the efficiency prediction model, and the model calculates and outputs the data processing rate, computing power utilization, and delay fluctuation rate within the target time interval. The data processing rate refers to the amount of data processed by the system per unit time, the computing power utilization rate refers to the ratio of the computing resources actually used by the system to the total computing resources, and the delay fluctuation rate refers to the fluctuation range of the communication delay per unit time.

[0080] The data processing rate, computing power utilization, and latency fluctuation values ​​are normalized, converting them to the same range to eliminate dimensional differences between the parameters. This normalization process yields a dynamic perception evaluation value that comprehensively reflects the system's ability to perceive and handle dynamic changes in the current scenario.

[0081] Example 3: When the context association analysis module performs a joint analysis on the context association effectiveness of the system, it first retrieves the quantification result of the abnormal tracking state of the system. The quantification result of the abnormal tracking state is the abnormal state fusion value obtained by the multimodal information acquisition module through the fusion calculation of the tracking abnormality contribution, the communication abnormality rate and the environmental interference evaluation value. For this abnormal state fusion value, a corresponding correction factor is set. The value of the correction factor is determined based on the degree of influence of the abnormal state on the system context association analysis. Different ranges of abnormal state fusion values ​​correspond to different correction factor values. The tracking impact correction value is calculated by multiplying the abnormal state fusion value and the correction factor. This value is used to adjust the impact of the abnormal state in subsequent analysis.

[0082] After obtaining the tracking impact correction value, it is normalized and calculated together with the feature evaluation benchmark value generated by the target feature extraction module and the dynamic perception evaluation value generated by the scene dynamic perception module. During the normalization process, the three values ​​are mapped to the value range of 0-1 respectively. The specific method is: for each value, its possible minimum value is subtracted from the value, and then divided by the difference between its maximum and minimum values, so as to eliminate the influence of different indicators due to different dimensions and value ranges. After normalization, three values ​​of the same magnitude are obtained. These three values ​​are weighted and summed according to the preset ratio to obtain the context-related evaluation value. The calculation formula is as follows:

[0083]

[0084] in, Represents the contextual evaluation value, Represents the normalized feature evaluation benchmark value, represents the normalized dynamic perception evaluation value, represents the normalized tracking influence correction value, 、 、 They represent the weight coefficients corresponding to the feature evaluation benchmark value, dynamic perception evaluation value, and tracking impact correction value, respectively, and .

[0085] The system has a pre-set contextual evaluation threshold. This threshold is determined based on the system's contextual performance data during normal operation and reflects the minimum performance level required to meet tracking requirements through edge processing. The calculated contextual evaluation value is compared with this threshold. If the contextual evaluation value is greater than or equal to the threshold, the system's contextual performance is high and the edge's processing power is sufficient to handle the current tracking task. An edge control signal is generated. If the contextual evaluation value is less than the threshold, the system's contextual performance is insufficient, making it difficult to guarantee tracking results solely through edge processing. Therefore, cloud-based computing resources are required for enhanced tracking. A cloud-based enhancement signal is generated.

[0086] After receiving edge control signals or cloud-based reinforcement signals, the tracking strategy generation module initiates multi-level optimization strategy analysis. When an edge control signal is detected, a tracking maintenance command is triggered. This tracking maintenance command contains rules for adjusting vehicle tracking parameters, dynamically adjusting the vehicle's tracking window size and feature matching accuracy based on these rules. The tracking window size is adjusted based on the target vehicle's speed and image resolution. When the vehicle is moving at high speed, the tracking window is appropriately increased to prevent the target from escaping the tracking range; when the vehicle is moving at low speed or stationary, the tracking window is appropriately reduced to reduce unnecessary computation. Feature matching accuracy is adjusted based on the clarity of the target features. When the target features are clear, the matching accuracy is increased to enhance tracking accuracy; when the target features are blurred, the matching accuracy is reduced to ensure tracking continuity. These adjustments generate tracking maintenance parameters, which contain specific values ​​such as the adjusted tracking window size and feature matching threshold.

[0087] When a cloud-based reinforcement signal is detected, the computing power allocation directive is triggered. This directive includes a plan for allocating system resources, based on which the system's local storage capacity and cloud computing tasks are dynamically planned. Local storage capacity is adjusted based on real-time data traffic and the access frequency of historically stored data. Low-frequency data is transferred to cloud storage, freeing up local storage space to accommodate new data storage needs; high-frequency data is retained locally to improve data access speeds. Cloud computing tasks are planned based on task complexity and real-time requirements. Tasks requiring significant computing resources but with low real-time requirements are allocated to the cloud, while simple tasks with high real-time requirements are retained locally. These plans generate computing power allocation parameters, which include the local storage capacity allocation ratio, a list of cloud and local task allocations, and other information.

[0088] Through the above process, the tracking strategy generation module can generate targeted tracking maintenance parameters and computing power allocation parameters according to different signal types, realize dynamic optimization of the vehicle target tracking process, and enable the system to adapt to different scenario conditions and performance requirements.

[0089] Example 4: The system includes a permission management module, which continuously monitors the system's tracking operation permissions in real time. The monitoring process starts with parsing the system's license data during operation control. The license data records the identity information of the operating subject, the authorized operation scope, and the validity period of the operation. By parsing these data, the permission type of the operating subject is obtained. The permission type is divided into multiple levels according to the different operation contents, covering different operation permissions such as viewing, modifying, and exporting tracking data; at the same time, the historical violation records of the operating subject are extracted. The historical violation records include records of unauthorized operations, illegal data exports, and other behaviors that occurred during past operations.

[0090] The parsed permission type is matched with the system's preset operation permission list. The operation permission list specifies in detail the permitted operation ranges corresponding to different operating entities. For example, system administrators can perform all operations, while ordinary users can only view some tracking data. By comparing the actual permission type of the operating entity with the permission range specified in the list, the permission type deviation is calculated. The deviation value reflects the degree of difference between the actual permission and the required permission. If the actual permission exceeds the range specified in the list, the deviation is positive; if the actual permission is less than the specified range, the deviation is negative.

[0091] The frequency of unauthorized operations and the total number of operations in historical violation records are counted, and the difference between the two is calculated to obtain the operational behavior abnormality index. The frequency of unauthorized operations refers to the number of operations performed by the operator that exceed their authority within the past period. The total number of operations refers to the total number of operations performed by the operator within the same period. The difference rate is calculated as the ratio of the frequency of unauthorized operations to the total number of operations. This index directly reflects the operator's level of compliance.

[0092] The permission type deviation degree and the operational behavior anomaly index are weighted and fused according to a preset ratio. The weighting ratio is set to comprehensively consider the role of the compliance of the permission type itself and the standardization of historical operational behavior in permission assessment. This fusion calculation generates a permission association assessment value, a quantitative indicator that comprehensively reflects the compliance of the operating entity's permissions and the standardization of its behavior. The value ranges from 0 to 1. The closer the value is to 0, the more normal the permission association status is, while the closer the value is to 1, the higher the permission risk.

[0093] When the context-related analysis module conducts a joint analysis of the system's context-related effectiveness, the permission-related evaluation value generated above is included in the analysis scope. Based on the previous analysis, the permission-related evaluation value is further combined to perform an operation permission constraint analysis on the context-related effectiveness. Specifically, the permission-related evaluation value is normalized together with the feature evaluation baseline value, the dynamic perception evaluation value, and the tracking impact correction value so that all indicators are in the same value range. After that, the context-related evaluation value is recalculated, and the calculation at this time will take into account the weight of the permission-related evaluation value. The size of the weight is determined according to the degree of influence of the operation permission on the tracking effectiveness in the current scenario.

[0094] This operational permission constraint analysis allows the results of contextual relevance analysis to not only reflect target characteristics, scene dynamics, and tracking anomalies, but also incorporate operational permission compliance factors. A high permission relevance assessment indicates a certain level of permission risk, potentially negatively impacting the system's contextual relevance performance. In this case, when generating edge control signals or cloud reinforcement signals, the signal strength is adjusted accordingly or permission constraints are added. A low permission relevance assessment indicates normal operational permissions, with minimal constraints on contextual relevance performance. Signal generation is primarily based on other evaluation metrics.

[0095] By incorporating operational authority factors into contextual correlation analysis, the system can consider both technical performance and operational security factors, making the generated edge control signals and cloud-enhanced signals more comprehensive. This ensures not only technical feasibility but also compliance and security of the operational process when optimizing tracking strategies.

[0096] Example 5: The system includes a dynamic adjustment module. Its core function is to adjust the resource allocation frequency for system operations and maintenance in real time based on the tracking maintenance parameters and computing power allocation parameters output by the tracking strategy generation module. This module then feeds the adjusted parameters back to the multimodal information acquisition module, forming a closed-loop optimization process. This process enables the system to continuously optimize resource allocation based on actual operating conditions to adapt to changing tracking requirements in different scenarios.

[0097] After receiving the tracking maintenance parameters, the dynamic adjustment module first parses the parameters to determine whether they involve adjusting the tracking window size. The tracking window size is a critical parameter that affects target tracking accuracy and system resource consumption. Its value change is directly related to the range and frequency of sensor data collection. When the tracking maintenance parameters include a tracking window size adjustment instruction, the dynamic adjustment module activates the corresponding resource configuration adjustment mechanism and simultaneously adjusts the sensor scheduling parameters in the resource configuration. Sensor scheduling parameters include sensor sampling frequency, operating hours, and data transmission interval. For example, when the tracking window increases, it means that a wider monitoring range needs to be covered. In this case, the system will increase the sampling frequency of sensors in the relevant area and shorten the data transmission interval to ensure that target dynamics within a larger range can be captured. At the same time, the operating hours of sensors in the core area of ​​the tracking window are appropriately extended to ensure continuous collection of critical data. If the tracking window decreases, the sampling frequency of sensors in non-core areas is correspondingly reduced, and the data transmission interval is extended. This reduces unnecessary resource consumption and concentrates the saved resources on monitoring the core area.

[0098] When the computing power allocation parameters received by the dynamic adjustment module involve the optimization of cloud computing tasks, another set of resource configuration adjustment processes will be triggered to synchronously adjust the network bandwidth allocation parameters in the resource configuration. Network bandwidth allocation parameters include the communication bandwidth ratio between the edge and the cloud, the transmission priority of different types of data, etc. Cloud computing task optimization usually involves the reallocation of tasks, such as transferring some complex computing tasks originally undertaken by the edge to the cloud, or vice versa. When more tasks need to be allocated to the cloud, the system will increase the communication bandwidth ratio between the edge and the cloud to ensure that a large amount of computing data can be quickly transmitted to the cloud; at the same time, a higher transmission priority is set for data related to the computing tasks to avoid processing delays caused by data congestion. If the number of cloud computing tasks decreases and the edge takes on more computing tasks, the communication bandwidth ratio between the edge and the cloud will be reduced, and more bandwidth resources will be allocated to communication between devices within the edge to ensure efficient and coordinated data processing at the edge.

[0099] After adjusting the sensor scheduling parameters or network bandwidth allocation parameters, the dynamic adjustment module packages and integrates the adjusted parameters to form a new resource allocation plan and inputs it into the multimodal information acquisition module. After receiving the new resource allocation parameters, the multimodal information acquisition module restarts the optimization and verification process. During this process, the multimodal information acquisition module collects multi-source data according to the new parameters, including visual feature data and motion parameter data collected by the adjusted sensors, as well as environmental context information transmitted under the new network bandwidth configuration. Subsequently, the system processes and analyzes the data in the target feature extraction module, scene dynamic perception module, context association analysis module, and tracking strategy generation module according to the normal workflow, generating new tracking maintenance parameters and computing power allocation parameters. The dynamic adjustment module receives these new parameters again and compares them with the previous parameters. If any discrepancies are found, further adjustments are made until the generated parameters stabilize, forming a closed loop of continuous optimization.

[0100] Through such a closed-loop optimization process, the system can continuously correct resource allocation based on parameter feedback during actual operation, so that tracking maintenance and computing power allocation always match the current scenario requirements, thereby maintaining stable and efficient vehicle target tracking and recognition capabilities in complex and changing environments.

[0101] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0102] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An adaptive context-aware vehicle target tracking and recognition system, characterized in that: include: The multimodal information acquisition module is used to collect the vehicle's visual feature data, motion parameter data, and environmental context information in multiple dimensions to obtain a multi-source scene data set. Based on the multi-source scene data set, the module quantitatively identifies abnormal tracking states and generates tracking warning signals. The generated tracking warning signals trigger the scene dynamic perception module and the context association analysis module. The target feature extraction module is used to extract the vehicle's contour feature parameters, color distribution features, and texture detail features, dynamically evaluate the target's trackability, and obtain a feature evaluation benchmark value; The scene dynamic perception module is used to monitor the system's real-time data traffic, local computing load, and communication delay parameters, and to predict and analyze the scene's dynamic perception performance to obtain a dynamic perception evaluation value. The context-related analysis module receives the feature evaluation baseline value and the dynamic perception evaluation value, performs a joint analysis on the system's context-related effectiveness, and generates edge control signals and cloud-side reinforcement signals; The tracking strategy generation module is used to receive edge control signals and cloud reinforcement signals, perform multi-level optimization strategy analysis, and generate vehicle tracking maintenance parameters and computing power allocation parameters; The quantitative identification of abnormal tracking status based on the scenario multi-source data set includes: By extracting the target ambiguity, color offset and contour integrity from the vehicle's visual feature data, the ambiguity value, color offset value and contour integrity value are obtained. The values ​​of the three are extracted and weighted to calculate the contribution of tracking anomaly. By extracting the signal interruption frequency, parameter matching degree and data packet loss rate from the system's sensor communication status parameters, the communication interruption value, parameter matching value and data packet loss value are obtained, and they are marked as tracking stability characteristic values. A tracking stability threshold is set, and the characteristic value is compared and analyzed with the threshold. When the characteristic value is less than the threshold, the sensor is marked as an abnormal sensor. The ratio of the number of abnormal sensors to the total number of sensors in the current system is counted to obtain the communication anomaly rate. At the same time, the light intensity change and the frequency of occlusion in the environmental context information are extracted, and dynamic simulation calculations are performed to obtain the environmental interference assessment value. The tracking anomaly contribution, communication anomaly rate, and environmental interference assessment values ​​are multiplied by their corresponding weight coefficients and added together to obtain an abnormal state fusion value, which is then compared with a preset abnormality threshold. If the fusion value is higher than the threshold, a tracking warning signal is generated. The dynamic evaluation of the traceability of the target includes: By analyzing the spatial distribution of the vehicle's contour feature parameters, the corresponding contour change distribution map and occlusion risk assessment map are generated; Extract the reference distribution map of the standard tracking vehicle profile from the system database, perform a topological comparison between the target vehicle's profile change distribution map and the reference distribution map, calculate the profile matching degree of the two, and perform normalization processing to obtain the target tracking potential index; Extract the occlusion occurrence frequency and feature loss ratio from the vehicle occlusion risk assessment map and mark them as occlusion feature values ​​respectively; Extract the standard occlusion risk threshold from the system database, calculate the difference between the occlusion feature value and the threshold, and obtain the traceability deviation; The target tracking potential index and the trackability deviation value are weighted and fused to obtain the feature evaluation benchmark value.

2. The adaptive context-aware vehicle target tracking and recognition system according to claim 1, characterized in that: The predictive analysis of the dynamic perception effectiveness of the scene includes: By extracting the number of image frames, motion parameters and log storage capacity from the real-time data flow of the system, a data flow parameter set is obtained; Extract historical tracking data of similar scenarios from the system database, build an efficiency prediction model based on a dynamic balance algorithm, input the traffic parameter set into the model, and output the data processing rate, computing power utilization, and latency fluctuation rate within the target time interval; The data processing rate, computing power utilization, and delay fluctuation rate are normalized and calculated to obtain the dynamic perception evaluation value.

3. The adaptive context-aware vehicle target tracking and recognition system according to claim 1, characterized in that: The joint analysis of the context-related effectiveness of the system includes: Retrieve the system's abnormal tracking state quantification results, set its correction factor, and obtain the tracking impact correction value through calculation and processing; Normalizing the feature evaluation benchmark value, the dynamic perception evaluation value, and the tracking impact correction value to obtain a contextual correlation evaluation value; Set the context association evaluation threshold. If the association evaluation value is greater than or equal to the threshold, an edge control signal is generated; if it is less than the threshold, a cloud reinforcement signal is generated.

4. The adaptive context-aware vehicle target tracking and recognition system according to claim 1, characterized in that: The multi-level optimization strategy analysis includes: If an edge control signal is captured, the tracking maintenance instruction is triggered, and the vehicle's tracking window size and feature matching accuracy are dynamically adjusted according to the instruction to generate tracking maintenance parameters; If a cloud reinforcement signal is captured, the computing power allocation instruction will be triggered. Based on the instruction, the system's local storage capacity and cloud computing tasks will be dynamically planned to generate computing power allocation parameters.

5. The adaptive context-aware vehicle target tracking and recognition system according to claim 1, characterized in that: Also includes: The permission control module is used to monitor the system's tracking operation permissions in real time, extract the operator's identification, permitted operation scope and illegal operation characteristics, and generate permission association evaluation values; The context association analysis module further combines the authority association evaluation value to perform an operation authority constraint analysis on the context association effectiveness.

6. The adaptive context-aware vehicle target tracking and recognition system according to claim 5, characterized in that: The real-time monitoring of the system's tracking operation permissions includes: By analyzing the license data of the system during operation control, the permission type and historical violation records of the operating subject are obtained; Match the permission type with the preset operation permission list and calculate the permission type deviation; Calculate the difference between the frequency of unauthorized operations and the total number of operations in historical violation records to obtain the abnormal operation behavior index; The permission type deviation degree and the operation behavior abnormality index are weighted and fused to generate the permission association evaluation value.

7. The adaptive context-aware vehicle target tracking and recognition system according to claim 1, characterized in that: Also includes: The dynamic adjustment module is used to adjust the resource configuration frequency of system operation and maintenance in real time based on the generated tracking maintenance parameters and computing power allocation parameters, and feed the adjusted parameters back to the multimodal information acquisition module to form a closed-loop optimization process.

8. The adaptive context-aware vehicle target tracking and recognition system according to claim 7, characterized in that: The adjustment process of the dynamic adjustment module includes: If the tracking maintenance parameters involve adjusting the tracking window size, the sensor scheduling parameters in the resource configuration are adjusted synchronously; If the computing power allocation parameters involve cloud computing task optimization, the network bandwidth allocation parameters in the resource configuration will be adjusted simultaneously; Input the adjusted parameters into the multimodal information acquisition module and restart the optimization and verification process.

Images