Specific target person behavior identification method fusing camera and UWBAOA base station

Abstract

The invention relates to the technical field of image recognition, in particular to a specific target person behavior recognition method fusing a camera and a UWBAOA base station, which comprises the following steps: based on target person position data of the camera and the UWBAOA base station, the target person position data comprises node information of a starting point, a transfer point, speed change and a stay position in a target person movement track; motion features are extracted, the transfer trend of the nodes is analyzed, the feature membership degree of each behavior node is obtained in combination with motion intensity data, and a fuzzy category affiliation result is generated by sorting and classifying the feature membership values. According to the method, the motion trail of the target person is analyzed, node information such as the starting point, the transfer point, the speed change and the stop position in the motion path is divided into a plurality of segments, and the fuzzy category affiliation result is generated through membership calculation and sorting, so that deep classification of the behavior pattern of the target person is formed; and the accuracy of behavior recognition is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of image recognition technology, and in particular to a method for identifying the behavior of a specific target person by integrating a camera and a UWBAOA base station. Background Art

[0002] The field of image recognition technology refers to the technology of identifying and extracting information from image or video data through computer vision technology. Its core lies in the use of artificial intelligence, deep learning algorithms, feature extraction and pattern recognition technologies to analyze specific objects, scenes, actions, etc. in image data. Through the automatic processing and understanding of images by machine vision systems, the digital and intelligent application of visual information is realized.

[0003] Among them, the specific target person behavior recognition by integrating camera and UWBAOA base station is a method of combining image data and radio data for behavior recognition. The method uses the data fusion technology of camera and ultra-wideband (UWB) AOA (angle of arrival) base station to track and analyze the behavior of specific target persons. It is mainly used in public places, smart security, behavior analysis and other scenarios. By accurately monitoring and identifying individual behaviors, it can achieve more accurate security prevention and control and behavior analysis, thereby providing technical support for security management and behavior prediction.

[0004] In the existing technology, behavior recognition mainly relies on the basic data of cameras and ultra-wideband base stations. The motion characteristics of the target person are analyzed by superimposing the image and radio data. However, in terms of information processing of each node in the motion trajectory, it is difficult to deeply analyze the detailed characteristics such as speed changes and direction turning angles between nodes, resulting in behavior recognition being limited to basic action classification. In addition, the existing methods do not fully consider the curvature changes and speed characteristics of the front and rear nodes when processing missing fragments in the motion trajectory, resulting in insufficient data integrity and trajectory coherence, affecting the coherence and accuracy of the overall recognition. The existing technology has limited judgment on potential behaviors such as the path deviation trend of the target person, which limits the system's ability to warn of potential abnormal behaviors. At the same time, the response to detecting the deviation of the target person is slow, and it is difficult to quantify the deviation risk in a timely manner, affecting the accuracy of safety prevention and control. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings in the prior art and propose a specific target person behavior recognition method that integrates a camera and a UWBAOA base station.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: a method for identifying the behavior of a specific target person by integrating a camera and a UWBAOA base station, comprising the following steps:

[0007] S1: Based on the target person's location data from the camera and UWB base station, including the node information of the starting point, transfer point, speed change and stop position in the target person's motion trajectory, the motion features are extracted, and the transfer trend of the nodes is analyzed. The feature membership of each behavior node is obtained in combination with the motion intensity data. By sorting and classifying the feature membership values, the fuzzy category attribution results are generated;

[0008] S2: Based on the motion features in the fuzzy category attribution results, the feature change frequency and similarity of adjacent nodes are analyzed, the transition probability between target personnel nodes is calculated, multiple simulation paths are generated according to the transition probability, and the node features in the paths are aggregated and analyzed, and a group of paths with the highest transition probability are selected to generate a behavior trend prediction path cluster;

[0009] S3: Based on the behavior trend prediction path cluster, the current behavior trajectory of the target person is compared with the corresponding node in the path cluster, the deviation degree of each node is quantified, and the deviation difference information between adjacent nodes is cumulatively analyzed to evaluate the deviation trend of the overall trajectory, determine the deviation direction, and generate the target trajectory deviation evaluation result;

[0010] S4: Extract the missing segments in the target motion trajectory captured by the camera and the UWB base station, complete the missing nodes in the target trajectory by interpolation fitting, and evaluate the overall motion trend and trajectory consistency after the repaired trajectory in combination with the target trajectory deviation assessment result to generate the target person behavior recognition result.

[0011] As a further solution of the present invention, the step of acquiring the transfer trend of the analysis node is specifically as follows:

[0012] S111: Based on the data from the camera and the UWB base station, the location information of the target person at the time node is captured, the starting point, transfer point, speed change point and stop position in the motion trajectory are recorded, and the nodes are distinguished, and multiple motion features such as speed change rate, direction turning angle, position displacement and stop time are extracted to obtain the motion feature data of each trajectory;

[0013] S112: Classifying and summarizing the motion features between nodes according to the motion feature data of each trajectory, and statistically summarizing the motion features according to the distribution information of speed change, direction change, total displacement and dwell time to generate a data set of motion patterns between nodes;

[0014] S113: using a formula according to the data set of the inter-node motion pattern;

[0015]

[0016] Calculate the transfer trend index I to obtain the transfer trend analysis results of the behavior change;

[0017] Among them, w1, w2, w3, w4 are the weight coefficients of motion features, corresponding to speed change rate, direction turning angle, displacement and stay time respectively, Δv is the speed change rate, θ is the direction turning angle, D is the displacement, T is the stay time, max(Δv), max(θ), max(D), max(T) are the maximum values ​​of each feature.

[0018] As a further solution of the present invention, the step of obtaining the characteristic membership of each behavior node is specifically as follows:

[0019] S121: According to the transfer trend analysis result of the behavior change, the formula is used:

[0020]

[0021] Calculate the membership μ of the i-th node i , get the characteristic membership information of the node;

[0022] Among them, S i is the motion intensity of the ith node, S j is the motion intensity of the jth node, n is the total number of nodes;

[0023] S122: Based on the characteristic membership information of the node, all nodes are sorted from large to small according to the membership, and the membership interval is defined according to historical data, and the nodes of the membership are classified into multiple matching degree categories to obtain a fuzzy category attribution result.

[0024] As a further solution of the present invention, the step of obtaining the transfer probability between target personnel nodes is specifically as follows:

[0025] S211: Based on the motion feature data in the fuzzy category attribution result, by extracting the change information of the speed change rate, direction turning angle, position displacement and dwell time of adjacent nodes, comparing the feature change frequency and similarity between adjacent nodes, marking the nodes that meet the set threshold conditions, and generating the node's coherent path segment data;

[0026] S212: According to the continuous path segment data of the node, the formula is used:

[0027]

[0028] Calculate the transition probability P(a→b) from node a to node b, and obtain the transition probability data between the target personnel nodes;

[0029] Among them, f ab is the frequency of feature change from node a to node b, s abis the feature similarity between node a and node b, s ak is the similarity between node a and node k, n is the total number of nodes, and k is the index variable of the node.

[0030] As a further solution of the present invention, the step of obtaining the node features in the path through aggregation analysis is specifically as follows:

[0031] S221: Generate multiple simulation paths by random sampling according to the transfer probability data between the target personnel nodes, analyze the motion characteristics in each path, extract the change trend of speed, direction and position, and perform feature aggregation analysis to generate motion feature data of the path;

[0032] S222: Based on the motion feature data of the paths, the paths are sorted and ordered, paths with stable motion features are selected, and the paths are aggregated according to feature similarity to generate a behavior trend prediction path cluster.

[0033] As a further solution of the present invention, the step of obtaining the quantified offset degree of each node is specifically as follows:

[0034] S311: Based on the real-time trajectory data collected by the camera and the UWBAOA base station, the current behavior trajectory of the target person is sorted into a time series, matched with the node data in the behavior trend prediction path cluster, and the time, position and direction information of the node are compared to generate a node matching result;

[0035] S312: According to the node matching result, the formula is used:

[0036]

[0037] Calculate the deviation degree Δp of the i-th node i , get the node offset degree information;

[0038] in, is the distance of the i-th node in the current behavior trajectory of the target person, is the reference distance corresponding to the i-th node in the path cluster.

[0039] As a further solution of the present invention, the step of obtaining the deviation trend of the overall trajectory is specifically as follows:

[0040] S321: Based on the node offset degree information, analyze the offset direction and offset amount of each pair of adjacent nodes, record the offset difference information, and generate an offset sequence data table;

[0041] S322: According to the offset sequence data table, the offset directions and amplitudes of all path segments are sorted out, each segment is compared with the nodes in the path cluster, and marked and classified according to the differences, the deviation trend and direction of the overall trajectory are determined, and the target trajectory deviation evaluation result is generated.

[0042] As a further solution of the present invention, the steps of obtaining the target person behavior recognition result are specifically as follows:

[0043] S411: Based on the target trajectory deviation evaluation result, identify the front and rear nodes of the missing segment in the trajectory, sort out the timestamp, position coordinates and curvature, direction change and speed feature information of the nodes, adjust the interpolation parameters to make the trend of the curve fitting consistent with the feature changes of the front and rear nodes, and perform interpolation fitting to generate a completed virtual node;

[0044] S412: Based on the completed virtual node, the formula is used:

[0045]

[0046] Calculate the estimated node coordinate vector within the missing segment at time t′ Get estimated node location information;

[0047] Where G0 = (x0, y0, z0) is the position coordinate of the known node before the missing segment, G1 = (x1, y1, z1) is the position coordinate of the known node after the missing segment, t′0 and t′1 are the timestamps of the known nodes before and after the missing segment, respectively, and t′ is the interpolation time, which is between t′0 and t′1;

[0048] S413: According to the estimated node position information, the completed trajectory is compared with the motion features of the previous and next nodes, the motion features of the deviation nodes are adjusted to match the overall trajectory trend, and the target person behavior recognition result is generated.

[0049] Compared with the prior art, the advantages and positive effects of the present invention are:

[0050] In the present invention, by analyzing the motion trajectory of the target person, the node information such as the starting point, transfer point, speed change and stop position in the motion path is divided into multiple segments, and the fuzzy category attribution results are generated by membership calculation and sorting, forming a deep classification of the target person's behavior pattern, so that the accuracy of behavior recognition is significantly improved. By calculating the change frequency and similarity of these motion characteristics, the transfer probability between nodes can be derived, and multiple simulation paths are generated based on this. The behavior trend prediction path cluster is constructed with the help of continuity and consistency analysis between paths, so as to effectively predict the behavior development trend of the target person and enhance the timeliness and reliability of behavior analysis. Then compare the node information in the behavior trend path cluster with the current behavior trajectory, quantify the node deviation, evaluate the deviation direction and degree of deviation, so that the system can quickly identify the deviation in the trajectory and quantify its risk. In addition, the missing segments of the trajectory are interpolated and fitted to ensure the integrity of the data and the coherence of the trajectory. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a schematic diagram of the main steps of the present invention;

[0052] Figure 2 A flow chart of analyzing the transfer trend of nodes in the present invention;

[0053] Figure 3 A flow chart for obtaining the characteristic membership of each behavior node in the present invention;

[0054] Figure 4 A flow chart showing the calculation of the transfer probability between target personnel nodes in the present invention;

[0055] Figure 5 A flow chart of the present invention for performing aggregate analysis on node features in a path;

[0056] Figure 6 A flow chart for quantifying the degree of deviation of each node in the present invention;

[0057] Figure 7 A flow chart for evaluating the deviation trend of the overall trajectory for the present invention;

[0058] Figure 8 The present invention is a flow chart for obtaining target person behavior recognition results. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0060] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0061] See also Figure 1 The present invention provides a technical solution: a method for identifying the behavior of a specific target person by integrating a camera and a UWBAOA base station, comprising the following steps:

[0062] S1: Based on the target personnel position data of the camera and UWB base station, including the node information of the starting point, transfer point, speed change and stop position in the target personnel's motion trajectory, the trajectory part between every two adjacent nodes is defined as a segment, and multiple motion features such as speed change rate, direction turning angle, position displacement and stay time are extracted. By statistically summarizing the motion features between the target personnel nodes, the transfer trend of the nodes is analyzed. Combined with the motion intensity data, the motion intensity data is the speed change amplitude and the total displacement, recursive membership calculation is performed on each behavior node in the trajectory to obtain the feature membership degree, and the fuzzy category attribution result is generated by sorting and classifying the feature membership values;

[0063] S2: Based on the motion features in the fuzzy category attribution results, according to the information of speed change rate, direction turning angle, position displacement and stay time in the motion features, by analyzing the feature change frequency and similarity of adjacent nodes, the transfer probability between the target personnel nodes is calculated. According to the transfer probability, multiple simulation paths are generated by random sampling, and the node features in the path are aggregated and analyzed. By analyzing the continuity and consistency of the path transfer, a group of paths with the highest transfer probability are selected to extract key features. The key features are speed change trend, direction stability or position offset amplitude, and a behavior trend prediction path cluster is generated;

[0064] S3: Predict the path cluster based on the behavior trend, call all the node information in the path cluster, compare the current behavior trajectory of the target person with the corresponding node in the path cluster, calculate the node deviation, combine the key features extracted from the path cluster, quantify the deviation degree of each node, and conduct cumulative analysis through the deviation difference information between adjacent nodes to evaluate the deviation trend of the overall trajectory, determine the deviation direction, and generate the target trajectory deviation evaluation result;

[0065] S4: Extract the missing segments in the target motion trajectory captured by the camera and UWB base station, perform interpolation fitting on the curvature change, direction angle and speed change of the nodes before and after the missing segments, obtain the data of the missing segment nodes through fitting calculation, complete the missing nodes in the target trajectory, and use the minimum error completion method to ensure the continuity of the trajectory. Combined with the target trajectory deviation evaluation result, evaluate the overall motion trend and trajectory consistency after the repaired trajectory, correct the behavioral characteristics of the deviated part, and generate the target person behavior recognition result;

[0066] The fuzzy category attribution results include the classification of different movement patterns, the membership evaluation of behavior patterns and the intensity index of each behavior pattern. The behavior trend prediction path cluster includes the probability rating of each path, the identification of key behavior nodes and the behavior consistency analysis of the path pattern. The target trajectory deviation evaluation results include the deviation measurement between nodes, the deviation trend analysis of the behavior trajectory and the quantitative evaluation of the deviation direction. The target person behavior identification results include the consistency verification of the behavior trajectory, the correction of the behavior characteristics of the deviated part and the confirmation of the behavior prediction.

[0067] See also Figure 2 ,The specific steps for obtaining the transfer trend of the analysis node are:

[0068] S111: Based on the data from the camera and the UWB base station, the location information of the target person at the time node is captured, the starting point, transfer point, speed change point and stop position in the motion trajectory are recorded, and the nodes are distinguished, and multiple motion features such as speed change rate, direction turning angle, position displacement and stop time are extracted to obtain the motion feature data of each trajectory;

[0069] The location and timestamp data of the target person are captured based on the camera and UWB base station respectively. The camera is used to track the activity trajectory of the person, and the UWB base station provides real-time positioning information to obtain the relative time between each node. Through the combination of the two, the starting point, transfer point and stay position node in the target person's movement trajectory are obtained. In order to mark each trajectory, the displacement data of every two adjacent nodes is defined as a segment. Specifically, the spatial coordinates are determined by processing the signal response time of the UWB base station, and the movement speed of each node is obtained. The direction information obtains the azimuth according to the change direction of the adjacent coordinates. The turning angle on the motion path is quantified by adopting the commonly used trigonometric calculation method (such as the direction vector method), and the displacement is obtained by the coordinate distance between adjacent nodes. The measurement of the stay time is recorded based on the time difference at the same coordinate position.

[0070] S112: Classifying and summarizing the motion features between nodes according to the motion feature data of each trajectory, and statistically summarizing the motion features according to the distribution information of speed change, direction change, total displacement and dwell time to generate a data set of motion patterns between nodes;

[0071] According to the multiple motion feature data extracted from the camera and UWB base station, such as speed change rate, steering angle, displacement and dwell time, the feature information between these nodes is statistically analyzed and summarized. First, all trajectory segments are numbered according to the order of the nodes, and the speed change rate between each node is calculated respectively. The speed change data in different trajectory segments are summarized by interval statistics, and the average and maximum values ​​of the speed of each segment are calculated. By comparing the differences in the speed of each segment, the motion intensity level of the entire trajectory is obtained. In addition, in the analysis of steering angle, the distribution law of steering nodes is obtained by calculating the steering angle change frequency in the trajectory. In the summary of displacement, the total distance moved by the personnel is obtained by accumulating the displacement of each segment. The data processing of the dwell time adopts the time interval statistics method to sum up the duration of multiple dwell nodes and identify the specific location of high-frequency dwell. The specific location of high-frequency dwell refers to the place where the personnel stay at certain nodes in the trajectory path more times and for a longer time. High-frequency dwell is the behavior of appearing at the same location multiple times in a short period of time, and the cumulative dwell time exceeds the set threshold. By identifying these locations, key areas of concern can be determined, such as rest areas, entrances or cross passages.

[0072] S113: Based on the data set of the inter-node motion pattern, a formula is used;

[0073]

[0074] Calculate the transfer trend index I to obtain the transfer trend analysis results of the behavior change;

[0075] Among them, I is used to evaluate the intensity of the change of the motion trend between nodes, and its value range is between 0 and 1, indicating the strength of the motion trend. w1, w2, w3, and w4 are weight coefficients of the motion characteristics, corresponding to the speed change rate, direction turning angle, displacement, and dwell time, respectively. They are set according to the analysis requirements and historical behavior data. The weights are determined by tools such as multivariate regression analysis in MATLAB, satisfying w1+w2+w3+w4=1. For example, if the target area in a certain scenario is the airport security checkpoint, more attention is paid to the target person's dwell time and speed change, because long-term stay or sudden deceleration may indicate potential risks. Therefore, the following weights can be assigned: w1=0.2, w2=0.2, w3=0.2, w4=0.4. For w1=0.2 (speed change rate): At the airport security checkpoint, although speed changes may exist, they are not the main analysis indicator, so a lower weight is given. If this weight is too high , the system may be too sensitive to misjudge a small speed change as abnormal behavior. For w2=0.2 (turning angle): Since the moving line of the airport security checkpoint is relatively fixed and the space for direction change is limited, the weight of the turning angle is also low. Unless the personnel frequently turn in a specific area, this part of the data has little impact on behavior recognition. For w3=0.2 (displacement): In the security check channel, the displacement distance of the target personnel is usually short, so it is not used as the main indicator. If the weight is set too high, normal traffic behavior may be misjudged as abnormal. For w4=0.4 (stay time): Stay time is an important indicator in the security check area. If someone stays in the security check area for too long, it may indicate a problem or abnormal behavior. Therefore, this feature is given the highest weight. Δv is the speed change rate, which indicates the degree of speed change between two adjacent nodes. The speed data is obtained through the camera and the UWB base station. The calculation formula is: v1 and v2 are the speeds of adjacent nodes, t1 and t2 are the corresponding timestamps. The speed data is collected through tools such as Python's OpenCV library and UWBAOABaseStation positioning. θ is the direction turning angle, which indicates the direction change of two adjacent trajectories. The motion trajectory of the target person is obtained through the camera, and the direction angle is calculated according to the coordinate changes of the adjacent nodes. and is the direction vector of two adjacent trajectories, which is calculated using Python's Numpy library. Arccos is the inverse function of the cosine function, which is often used to inversely calculate the angle between two vectors when the cosine value of the vector angle is known. D is the displacement, which represents the spatial distance between the two nodes. The displacement is calculated based on the positioning data of the UWB base station: (x1, y1) and (x2, y2) are the coordinates of adjacent nodes. T is the dwell time, which indicates the time a person stays at a specific node in seconds. It is calculated by the difference between the inter-frame timestamp of the camera and the timestamp recorded by the UWB base station: T = t2-t1. max(Δv), max(θ), max(D), max(T) are the maximum values ​​of each feature, which are used for normalization. These maximum values ​​are obtained based on historical data and stored in the database, which are extracted through SQL queries.

[0076] According to historical data, the following parameters can be obtained: Δv = 0.4m / s 2 , θ=45°, D=10m, T=8s; the corresponding maximum values ​​are: max(Δv)=1m / s 2 , max(θ)=90°, max(D)=15m, max(T)=10s; the determined weights are as follows: w1=0.3, W2=0.2, W3=0.3, W4=0.2.

[0077] Enter the formula to calculate:

[0078]

[0079] Calculate the contribution of each term:

[0080] I=0.3·0.4+0.2·0.5+0.3·0.6667+0.2·0.8

[0081] I = 0.12 + 0.1 + 0.2 + 0.16 = 0.58

[0082] The result I = 0.58 indicates that the movement trend of this segment of the trajectory is at a medium to high level on the normalized scale. If the system's baseline value is 0.5, this segment of the trajectory reflects a more active movement pattern, indicating that the target person frequently moves in this segment of the trajectory and changes direction greatly. This indicator can be used to determine the behavioral characteristics in a specific area.

[0083] See also Figure 3 , the specific steps for obtaining the feature membership of each behavior node are:

[0084] S121: Based on the results of the transfer trend analysis of behavioral changes, the formula is used:

[0085]

[0086] Calculate the membership μ of the i-th node i , get the characteristic membership information of the node;

[0087] Among them, μ iThe value range is between 0 and 1, indicating the degree of matching of the behavior of the node in the entire trajectory. i is the motion intensity of the ith node, calculated from the velocity change rate and displacement: S i =Δv i +D i , Δv i is the velocity change rate of the ith node, D i is the displacement of the ith node, n is the total number of nodes, and represents the number of all monitored behavior nodes in the trajectory. j is the motion intensity of the jth node.

[0088] For example, there are 3 nodes, and their motion intensities are: S1=5.2, S2=3.5, S3=7.

[0089] Calculate the membership of the first node S1:

[0090]

[0091] First calculate the square of the ratio:

[0092]

[0093] Substituting the values ​​into the formula:

[0094]

[0095] The results show that the calculated membership μ1≈0.265.

[0096] S122: based on the characteristic membership information of the nodes, all nodes are sorted from large to small according to the membership, and the membership interval is defined according to the historical data, and the nodes of the membership are classified into multiple matching degree categories to obtain the fuzzy category attribution result;

[0097] According to the calculated characteristic membership μ1=0.265, the system needs to sort and classify the membership values ​​of all nodes. First, the membership values ​​of each node are extracted through the SQL database, and the membership values ​​are arranged in descending order using the quick sort algorithm. During the sorting process, the system compares the membership of adjacent nodes one by one. If the current node μ1=0.265 is in a lower matching interval, the sorting algorithm places it at the end of the list. According to the classification standard of the system, the interval of membership values ​​is divided into: 0.7-1 is a high matching degree, 0.4-0.7 is a medium matching degree, and 0-0.4 is a low matching degree. Node μ1=0.265 belongs to the "low matching degree" category, and the system marks the node with a potential risk label, indicating that there may be an abnormality in this segment of behavior. In actual operation, the system will analyze multiple consecutive "low matching degree" nodes. If multiple nodes with similar low μ values ​​are found to appear continuously, the alarm mechanism will be triggered, indicating that the behavior of the target person needs to be further tracked and investigated. After the classification is completed, the system automatically stores the results in the database and assigns corresponding category labels to each node, such as node 1 is marked as "low match". In addition, the corresponding monitoring and early warning strategies are triggered based on the data of these labels. For example, if the membership of three consecutive nodes is in the range of 0.00.4, the system will generate an early warning notification for further inspection and response by security managers. At the same time, the classification results are also used to analyze the overall movement trend. By counting the distribution of nodes of different categories in the trajectory, it is determined whether there is abnormal behavior in the movement path. Finally, all data are collated to generate behavior fuzzy category attribution results, which are provided to managers as a reference for decision support.

[0098] See also Figure 4 , the specific steps for calculating the transition probability between target personnel nodes are:

[0099] S211: Based on the motion feature data in the fuzzy category attribution result, by extracting the change information of the speed change rate, direction turning angle, position displacement and dwell time of adjacent nodes, comparing the feature change frequency and similarity between adjacent nodes, marking the nodes that meet the set threshold conditions, and generating the node's coherent path segment data;

[0100] First, based on the motion feature data in the fuzzy category attribution results, the system screens and analyzes the feature change frequency and similarity of adjacent nodes from the dimensions of speed change rate, direction turning angle, position displacement and residence time. These analysis results will be used as the basic data for calculating the transfer probability between target personnel nodes. First, the system extracts the motion features of each pair of adjacent nodes from the historical trajectory database, and uses SQL query to summarize the change frequency of each feature, such as the frequency of speed change times exceeding the specified threshold, the frequency of direction angle deflection exceeding the set range, etc. Then, the system performs similarity analysis on the node features. For the comparison of direction turning angles, the system uses cosine similarity to measure the similarity of the angle between two direction vectors; for the residence time and position displacement, the dynamic time warping algorithm (DTW) is used to evaluate the degree of trajectory matching; the similarity of speed change rate is judged by setting a change threshold. If the speed change value between adjacent nodes is within the threshold, it is marked as high similarity. Finally, the frequency and similarity data are stored in the database and combined and analyzed in different feature dimensions. When the feature change frequency of multiple consecutive nodes is high and the similarity meets certain conditions, the system marks these nodes as coherent path segments. The frequency and similarity data of these path segments will be used as input to calculate the transition probability between nodes to ensure that the calculation of the transition probability is more in line with the historical movement law.

[0101] S212: Based on the node's coherent path segment data, use the formula:

[0102]

[0103] Calculate the transition probability P(a→b) from node a to node b, and obtain the transition probability data between the target personnel nodes;

[0104] Among them, P(a→b) is used to reflect the possibility of the target person transferring from node a to node b, and f ab is the frequency of feature changes from node a to node b, indicating the number of times this transition occurred in the historical trajectory. These data are obtained through SQL database query. ab is the feature similarity between node a and node b, with a value between 0 and 1, indicating the similarity of the motion features of the two nodes. The similarity of the direction angle is calculated using the cosine similarity algorithm, while the dwell time and position displacement are evaluated using the dynamic time warping algorithm (DTW). ak is the similarity between node a and node k, n is the total number of nodes, k is the index variable of the node, which is used to traverse all possible target nodes transferred from node a. Its value range is 1 to n, representing each node in the set of all nodes to which the target person may transfer.

[0105] For example, the system analyzes the transfer path segment of node a and obtains the following data: node a→b1: Node a→b2: Node a→b3: The total number of nodes n=3.

[0106] Calculate the product of frequency and similarity for each path:

[0107]

[0108]

[0109]

[0110] Compute the weighted sum of all paths:

[0111]

[0112] Calculate the transition probability of node a→b1:

[0113]

[0114] Calculate the transition probability of node a→b2:

[0115]

[0116] Calculate the transition probability of node a→b3:

[0117]

[0118] The result shows that the transition probability from node a to node b1 is 0.573, indicating that the characteristics of this path in the historical data change frequently and have a high similarity, and it is one of the preferred paths of the system. In contrast, the transition probabilities of nodes a→b2 and a→b3 are lower, indicating that the coherence or historical tendency of these paths is weaker.

[0119] See also Figure 5 , the specific steps for obtaining the node features in the path for aggregation analysis are:

[0120] S221: Generate multiple simulation paths by random sampling based on the transfer probability data between target personnel nodes, analyze the motion characteristics in each path, extract the change trend of speed, direction and position, and perform feature aggregation analysis to generate motion feature data of the path;

[0121] Based on the transition probability, multiple simulation paths are generated using random sampling, and the node features in the path are aggregated and analyzed. First, the calculated transition probability P(a→b) is organized into a probability matrix according to the node pairs to ensure that the probability data of each path is ready. Then, starting from the initial node, the system assigns weights according to the transition probability of each node, and performs random sampling based on probability at each transition. This process is repeated until a complete path is generated to ensure that the transition of all nodes in the path conforms to the historical motion pattern. In the multiple paths generated, it is necessary to extract the key motion features of the nodes, and analyze the speed, direction change, and displacement of each path segment by segment. The system traverses each pair of adjacent nodes in the path, gradually summarizes the speed change amplitude, direction turning angle, and dwell time data of these nodes, and aggregates these feature information. For example, for speed changes, it is necessary to accumulate the speed differences between all nodes to generate an overall speed change trend curve. The aggregated path data is stored in a table or database for further analysis of the continuity and consistency of the path. By comparing each path one by one, the system marks the paths that do not meet the coherence standard and selects the preferred path segments that meet the standard for subsequent processing.

[0122] S222: sorting and arranging the paths based on the motion feature data of the paths, selecting paths with stable motion features, and aggregating the paths according to feature similarity to generate a behavior trend prediction path cluster;

[0123] First, the generated paths are sorted and sorted. According to the weight values ​​of each path in the transfer probability matrix, the path with the highest probability is placed first, and other high-probability paths are gradually screened out to ensure that the analysis focuses on the most representative path segments. After the path is selected, the system processes the motion characteristics of each segment in the path. First, the speed change trend in each path is extracted, all speed data in the path are arranged in chronological order, and the acceleration and deceleration stages in the path are identified. For the analysis of direction changes, the system obtains the direction offset information of each segment of the path and marks the parts with frequent or stable direction changes to determine whether the direction of the movement is coherent. In the analysis of position offset, the displacement between nodes needs to be calculated segment by segment according to the node order of the path, and the offset trend is sorted out. Subsequently, these speed, direction and displacement feature data are integrated to form the motion feature curve of the path. The system compares the feature curves of all paths and screens out the path segments that meet the feature stability requirements. In the process of aggregating these preferred paths into prediction path clusters, the system first standardizes the motion features of each path to ensure that the feature data of different paths are compared under the same dimension. Specifically, the system calculates the average speed, average direction, and average displacement of each path, and normalizes these features to a range of 0 to 1 to eliminate the impact of dimensional differences. Next, the system uses a clustering algorithm (such as Kmeans or DBSCAN) to perform cluster analysis on the normalized path features. During the clustering process, the system first calculates the distance matrix based on the similarity of the path features, and then divides the paths into different clusters based on the set similarity threshold. Each cluster represents a group of paths with similar motion patterns, and the system automatically determines the central path of each cluster as the representative path of the cluster. In addition, the system also calculates the cohesion and separation of each path cluster to evaluate the stability and representativeness of the path cluster. The cohesion of the path cluster reflects the similarity between the paths, and the separation evaluates the differences between different path clusters. Only when the cohesion is high and the separation is significant, the path cluster will be marked as a valid predicted path cluster.

[0124] See also Figure 6 ,The specific steps for obtaining the quantified offset degree of each node are:

[0125] S311: Based on the real-time trajectory data collected by the camera and the UWBAOA base station, the current behavior trajectory of the target person is sorted into a time series, matched with the node data in the behavior trend prediction path cluster, and the time, location and direction information of the node are compared to generate a node matching result;

[0126] All node information in the path cluster is called to associate the current behavior trajectory of the target person with the corresponding node in the path cluster. First, the real-time data obtained by the camera and the UWBAOA base station are synchronously processed to convert the current position of the target person into a standardized three-dimensional coordinate point (x, y, z), and the timestamp is recorded to ensure the consistency of the time series in the path. The system automatically generates a node list based on the collected trajectory data, marking each node as T1, T2, etc. to ensure the consistency of the time sequence. Then, the system searches for the historical data of these nodes through the node information library in the path cluster. The operation steps include matching the nodes in the current trajectory with the nodes in the path cluster one by one, and automatically matching according to the similarity of spatial position, time interval and direction information. To reduce errors, the system sets a threshold range (such as a displacement difference within 0.5 meters) to ensure that the two nodes are considered to be matched successfully only when their positions are close enough. If the match fails, it automatically jumps to the next possible matching node for retry. During the matching process, the system not only compares the spatial position, but also checks the direction change, and analyzes whether the direction offset angle between the nodes is consistent with the historical pattern of the path cluster. For example, for the steering angle between node A and node B, the system will extract the direction vectors of the two to ensure that their error is within a reasonable range. After the matching is completed, the system marks the successfully associated nodes as paired nodes and saves the matched node information to the trajectory matching table to obtain the node matching results.

[0127] S312: According to the node matching result, the formula is used:

[0128]

[0129] Calculate the deviation degree Δp of the i-th node i , get the node offset degree information;

[0130] Among them, Δp i Represents the relative difference between the current behavior trajectory and the reference node in the path cluster, which is used to evaluate the deviation of the target person at a specific node. is the distance of the i-th node in the current behavior trajectory of the target person. This distance is obtained by fusing the data of the camera and the UWBAOA base station. is the reference distance corresponding to the ith node in the path cluster, which comes from the path cluster in the historical behavior data and represents the expected distance at the same location. The data is extracted from the existing trajectory database by the system. It represents the absolute difference between the current node distance and the reference node distance, which is used to eliminate the impact of positive and negative differences.

[0131] If the distance of the i-th node of the current behavior trajectory Reference node distance of path cluster

[0132] Calculate the degree of deviation:

[0133]

[0134] The result shows that at this node, the target person's behavioral trajectory distance is offset by 0.5 (i.e. 50%) from the expected distance of the reference path. By accumulating the degree of deviation of multiple nodes, it is possible to determine whether the entire trajectory is significantly different from the expected path, and further analyze the overall trend and direction of trajectory deviation. This analysis process provides a quantitative basis for real-time behavior monitoring and abnormal behavior warning.

[0135] See also Figure 7 ,The specific steps for obtaining the deviation trend of the overall trajectory are:

[0136] S321: Based on the node offset degree information, analyze the offset direction and offset amount of each pair of adjacent nodes, record the offset difference information, and generate an offset sequence data table;

[0137] The system organizes each trajectory into a sequence in chronological order and generates corresponding spatial offset data. On the basis of the completed preliminary node matching, the system analyzes the offset differences of adjacent nodes in timestamp order and calculates the spatial position changes of each pair of adjacent nodes in turn. During the operation, the offset direction and offset between each two nodes are extracted one by one to form a complete offset sequence. During the processing, the system will mark the changing trends of all offset directions, including the trend of continuous increase or decrease in offset. At the same time, the offset sequence of each pair of adjacent nodes is smoothed to eliminate possible positioning errors or noise data to ensure the continuity and accuracy of the trajectory data. The system further stores the cumulative offset values ​​between each adjacent node as a time series data table, and sets different offset thresholds for segmented analysis.

[0138] S322: According to the offset sequence data table, the offset directions and amplitudes of all path segments are sorted, each segment is compared with the nodes in the path cluster, and marked and classified according to the differences, the deviation trend and direction of the overall trajectory are determined, and the target trajectory deviation assessment result is generated;

[0139] First, the system checks the deviation direction of all path segments and compares it with the node information in the standard path cluster. In specific operations, each segment of deviation data will be matched with the corresponding segment in the path cluster. If the deviation direction is consistent with the standard direction but the deviation amplitude is large, the system will further mark it as "mild deviation"; if the deviation direction and amplitude are both significantly different from the standard path cluster, it will be marked as "serious deviation". Based on these marked data, the system generates a chart and analysis report of the cumulative deviation trend, and comprehensively analyzes the deviation direction of all path segments to determine the degree of deviation of the overall motion trajectory. In addition, the system will set the risk level: for example, a slight deviation will be classified as a "warning", while a serious deviation will trigger a real-time warning and output a deviation trend report through the monitoring platform. In the final step, the system generates a target trajectory deviation assessment result based on the overall trajectory deviation data to provide subsequent decision support for relevant personnel and the monitoring system.

[0140] See also Figure 8 , the specific steps for obtaining the target person behavior recognition results are:

[0141] S411: Based on the target trajectory deviation evaluation result, identify the front and rear nodes of the missing segment in the trajectory, sort out the timestamp, position coordinates and curvature, direction change and speed feature information of the nodes, adjust the interpolation parameters to make the trend of the curve fitting consistent with the feature changes of the front and rear nodes, and perform interpolation fitting to generate the completed virtual nodes;

[0142] First, the front and rear nodes of the missing fragment are identified to determine the changing trends of their positions, speeds, direction angles, and curvatures. The timestamps, spatial coordinates, and motion feature information of these nodes are obtained from the historical data of the camera and UWB base station, and are sorted in chronological order to ensure the continuity between the front and rear nodes. During the interpolation fitting process, the system selects a suitable interpolation method (such as polynomial interpolation or spline interpolation) based on the motion mode of the known nodes. When adjusting the interpolation parameters, the shape of the interpolation curve is matched with the changes in speed and direction based on the characteristic change trend of the front and rear nodes. Specifically, it includes: selecting the order of the fitting curve so that its smoothness conforms to the trend change of the node characteristics. In spline interpolation, the position of the control point is adjusted so that the virtual node after fitting conforms to the direction and speed changes of the front and rear nodes. The fitting interval step size is dynamically adjusted to keep the spacing of the fitting nodes at a reasonable distance from the front and rear nodes and avoid abrupt changes. The virtual node after completion is finally generated and is consistent with the known front and rear nodes. The system aligns the fitted trajectory with the motion features of the front and rear nodes to ensure that the trajectory is continuous and conforms to the law of motion.

[0143] S412: Based on the completed virtual node, the formula is:

[0144]

[0145] Calculate the estimated node coordinate vector within the missing segment at time t′ Get estimated node location information;

[0146] in, Used to complete the position of the missing node, G0 = (x0, y0, z0) is the position coordinate of the known node before the missing segment, indicating the target position coordinate detected by the camera or UWB base station at the previous moment. The coordinate unit is meter (m), which is obtained by fusing the camera visual data and UWB positioning data. G1 = (x1, y1, z1) is the position coordinate of the known node after the missing segment. Similar to G0, it indicates the target position at the next moment. The data is collected by the device and transmitted to the database for processing. t′0 and t′1 are the timestamps of the known nodes before and after the missing segment, respectively. The format is generally seconds (s) or milliseconds (ms). It is obtained through camera timestamp and UWB time synchronization to ensure device time consistency. t′ is the interpolation time, which is between t′0 and t′1, indicating the specific time node that needs to be completed. The time synchronization algorithm is used to ensure data alignment between the camera and the UWB base station.

[0147] If the previous node position is: G0 = (4, 3), time t′0 = 0 seconds; the next node position is: G1 = (10, 7), time t′1 = 5 seconds; the target interpolation time is: t′ = 2 seconds.

[0148] Substituting into the formula:

[0149]

[0150] Calculate weights:

[0151]

[0152] Calculate item by item:

[0153]

[0154]

[0155] The results show that at t′=2 seconds, the node position obtained by interpolation calculation is (6.4, 4.6), which reasonably completes the missing part of the trajectory and maintains the continuity of the trajectory in space and time.

[0156] S413: According to the estimated node position information, the completed trajectory is compared with the motion characteristics of the previous and next nodes, the motion characteristics of the deviation nodes are adjusted to match the overall trajectory trend, and the target person behavior recognition result is generated;

[0157] On the completed trajectory, the system uses the minimum error completion method to ensure the continuity and logic of the trajectory. First, the system identifies the nodes in the completed trajectory that do not meet expectations based on the target trajectory deviation assessment results. In this process, the system obtains the speed difference, direction angle difference and position change data between adjacent nodes, and organizes these errors into a matrix form, specifically including: taking the speed, direction and position deviation as the column data in the matrix respectively, and the rows of the matrix correspond to the node information at each time point. Different types of errors are normalized to have the same dimension to avoid analysis deviations caused by unit differences. The data in the matrix is ​​analyzed, the nodes whose errors exceed the set threshold are marked, and these nodes are compared with the motion characteristics of the previous and next nodes. During the comparison process, the system gradually adjusts the motion parameters of the marked nodes, such as direction and speed, to make them consistent with the overall trajectory trend and avoid the existence of local anomalies. Finally, the system generates a repaired and completed trajectory, verifies the consistency with the original trajectory, and organizes the analysis results into a detailed report as the basis for target personnel behavior recognition.

[0158] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A method for identifying specific target personnel behavior by integrating a camera and a UWBAOA base station, characterized in that: The following steps are involved: Based on the target person's location data from the camera and UWB base station, including the node information of the starting point, transfer point, speed change and stop position in the target person's motion trajectory, the motion features are extracted, and the transfer trend of the nodes is analyzed. The feature membership of each behavior node is obtained in combination with the motion intensity data. By sorting and classifying the feature membership values, the fuzzy category attribution results are generated. Based on the motion features in the fuzzy category attribution results, the feature change frequency and similarity of adjacent nodes are analyzed, the transition probability between target personnel nodes is calculated, multiple simulation paths are generated according to the transition probability, and the node features in the paths are aggregated and analyzed, and a group of paths with the highest transition probability are selected to generate a behavior trend prediction path cluster; Based on the behavior trend prediction path cluster, the current behavior trajectory of the target person is compared with the corresponding node in the path cluster, the deviation degree of each node is quantified, and the deviation difference information between adjacent nodes is cumulatively analyzed to evaluate the deviation trend of the overall trajectory, determine the deviation direction, and generate the target trajectory deviation evaluation result; The missing segments in the target motion trajectory captured by the camera and UWB base station are extracted, and the missing nodes in the target trajectory are completed by interpolation fitting. Combined with the target trajectory deviation assessment result, the overall motion trend and trajectory consistency after the repaired trajectory are evaluated to generate the target person behavior recognition result.

2. The specific target person behavior recognition method of the fusion camera and UWBAOA base station according to claim 1 is characterized in that: The steps for obtaining the transfer trend of the analysis node are specifically as follows: Based on the data from the camera and UWB base station, the location information of the target person at the time node is captured, the starting point, transfer point, speed change point and stop position in the motion trajectory are recorded, and the nodes are distinguished. Multiple motion features such as speed change rate, direction turning angle, position displacement and stop time are extracted to obtain the motion feature data of each trajectory; According to the motion feature data of each trajectory, the motion features between nodes are classified and summarized, and the motion features are statistically summarized according to the distribution information of speed change, direction change, total displacement and dwell time to generate a data set of motion patterns between nodes; Based on the data set of the inter-node motion pattern, the formula is adopted; Calculate the transfer trend index I to obtain the transfer trend analysis results of the behavior change; Among them, w1, w2, w3, w4 are the weight coefficients of motion features, corresponding to speed change rate, direction turning angle, displacement and stay time respectively, Δv is the speed change rate, θ is the direction turning angle, D is the displacement, T is the stay time, max(Δv), max(θ), max(D), max(T) are the maximum values ​​of each feature.

3. The specific target person behavior recognition method of the fusion camera and UWBAOA base station according to claim 2 is characterized in that: The steps for obtaining the characteristic membership of each behavior node are specifically as follows: According to the transfer trend analysis results of the behavior changes, the formula is used: Calculate the membership μ of the i-th node i , get the characteristic membership information of the node; Among them, S i is the motion intensity of the ith node, S j is the motion intensity of the jth node, n is the total number of nodes; Based on the characteristic membership information of the nodes, all nodes are sorted from large to small according to membership, and membership intervals are defined according to historical data, and the nodes of membership are classified into multiple matching degree categories to obtain fuzzy category attribution results.

4. The specific target person behavior recognition method of the fusion camera and UWBAOA base station according to claim 3 is characterized in that: The steps for obtaining the transfer probability between target personnel nodes are specifically as follows: Based on the motion feature data in the fuzzy category attribution result, by extracting the change information of the speed change rate, direction turning angle, position displacement and dwell time of adjacent nodes, comparing the feature change frequency and similarity between adjacent nodes, marking the nodes that meet the set threshold conditions, and generating the node's coherent path segment data; According to the coherent path segment data of the node, the formula is adopted: Calculate the transition probability P(a→b) from node a to node b, and obtain the transition probability data between the target personnel nodes; Among them, f ab is the frequency of feature change from node a to node b, s ab is the feature similarity between node a and node b, s ak is the similarity between node a and node k, n is the total number of nodes, and k is the index variable of the node.

5. The specific target person behavior recognition method of the fusion camera and UWBAOA base station according to claim 4 is characterized in that: The steps of obtaining the node features in the path through aggregation analysis are specifically as follows: According to the transfer probability data between the target personnel nodes, multiple simulation paths are generated by random sampling, the motion characteristics in each path are analyzed, the change trend of speed, direction and position is extracted, and feature aggregation analysis is performed to generate motion feature data of the path; Based on the motion feature data of the paths, the paths are sorted and ordered, paths with stable motion features are selected, and the paths are aggregated according to feature similarity to generate a behavior trend prediction path cluster.

6. The method for identifying specific target person behavior by integrating a camera and a UWBAOA base station according to claim 5, characterized in that: The steps for obtaining the quantified deviation degree of each node are specifically as follows: Based on the real-time trajectory data collected by the camera and the UWBAOA base station, the current behavior trajectory of the target person is sorted into a time series, matched with the node data in the behavior trend prediction path cluster, and the time, location and direction information of the node are compared to generate a node matching result; According to the node matching results, the formula is adopted: Calculate the deviation degree Δp of the i-th node i , get the node offset degree information; in, is the distance of the i-th node in the current behavior trajectory of the target person, is the reference distance corresponding to the i-th node in the path cluster.

7. The method for identifying specific target person behavior by integrating a camera and a UWBAOA base station according to claim 6 is characterized in that: The steps for obtaining the deviation trend of the overall trajectory are specifically as follows: Based on the node offset degree information, analyzing the offset direction and offset amount of each pair of adjacent nodes, recording the offset difference information, and generating an offset sequence data table; According to the offset sequence data table, the offset directions and amplitudes of all path segments are sorted out, each segment is compared with the nodes in the path cluster, and marked and classified according to the differences, the deviation trend and direction of the overall trajectory are determined, and the target trajectory deviation evaluation result is generated.

8. The method for identifying specific target person behavior by integrating a camera and a UWBAOA base station according to claim 7, characterized in that: The steps for obtaining the target person behavior recognition result are specifically as follows: Based on the target trajectory deviation assessment result, identify the front and rear nodes of the missing segment in the trajectory, sort out the timestamp, position coordinates and curvature, direction change and speed feature information of the nodes, adjust the interpolation parameters to make the trend of the curve fitting consistent with the feature changes of the front and rear nodes, and perform interpolation fitting to generate the completed virtual nodes; Based on the completed virtual node, the formula is adopted: Calculate the estimated node coordinate vector within the missing segment at time t′ Get estimated node location information; Where G0=x0, y0, z0 is the position coordinate of the known node before the missing segment, G1=x1, y1, z1 is the position coordinate of the known node after the missing segment, t′0 and t′1 are the timestamps of the known nodes before and after the missing segment, respectively, and t′ is the interpolation time, which is between t′0 and t′1; According to the estimated node position information, the completed trajectory is compared with the motion features of the previous and next nodes, the motion features of the deviation nodes are adjusted to match the overall trajectory trend, and the target person behavior recognition result is generated.

Images