Suspicious behavior recognition method and system based on pedestrian trajectory and stay analysis
By constructing a dynamic radial constant distance and dynamic time warping algorithm, combined with unidirectional rotation ratio and morphological deviation, a surround behavior score is generated, which solves the problem in the existing technology of being unable to distinguish between pedestrian surround behavior and straight-line passing behavior, and achieves highly accurate identification of suspicious behavior.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HENAN POLICE ACAD
- Filing Date
- 2026-04-29
- Publication Date
- 2026-07-14
Smart Images
Figure CN122392097A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of image recognition technology, specifically to a method and system for identifying suspicious behavior based on pedestrian trajectory and dwelling analysis. Background Technology
[0002] In the field of security monitoring of critical infrastructure, identifying abnormal reconnaissance behavior at fixed facilities (such as entrances / exits, guard posts, and warehouse doors) is a key aspect of security protection. Unlike ordinary pedestrians, personnel conducting reconnaissance or surveillance typically move laterally around the target facility to obtain visual information from different angles while maintaining a certain observation distance.
[0003] Existing technologies largely rely on simple statistical features such as area intrusion or speed and dwell time, making it difficult to distinguish between similar traversing and circling behaviors in polar coordinate scenarios. Theoretically, the radial distances of the two are quite different, but pauses and speed changes in actual movement can cause temporal redundancy and nonlinear fluctuations in the measured trajectory. Traditional point-to-point comparison methods, which require strict data alignment, cannot tolerate such temporal jitter, making it difficult for the system to effectively distinguish between circling accompanied by pauses and straight-line passage, easily leading to missed or false alarms, thus reducing the accuracy of identifying suspicious pedestrian behavior. Summary of the Invention
[0004] To address the aforementioned technical problems, the purpose of this application is to provide a method and system for identifying suspicious behavior based on pedestrian trajectory and dwell time analysis. The specific technical solution adopted is as follows: In a first aspect, embodiments of this application provide a method for identifying suspicious behavior based on pedestrian trajectory and dwelling analysis, the method comprising the following steps: Real-time acquisition of the measured distance and cumulative rotation angle from the target to the center point of the protected facility within the monitoring area; Within the preset current time window, the radial constant distance is calculated based on the measured distance from the target to the center point of the protected facility, which is used to characterize the radial characteristics of the ideal circular motion; a distance matrix is constructed between the measured distance and the radial constant distance, and the optimal warping path is searched in the distance matrix using a dynamic time warping algorithm to determine the original shape deviation value of the target within the current time window, which is used to characterize the geometric deviation of the measured distance relative to the radial constant distance of the ideal circular motion; Based on the net rotation amplitude of the cumulative rotation angle within the current time window and the difference in cumulative rotation angle between all adjacent times, the unidirectional rotation ratio is calculated and combined with the original shape deviation to generate the target's orbital behavior score within the current time window, which is used to characterize the comprehensive suspiciousness of the target that simultaneously satisfies the two characteristics of continuous unidirectional rotation and ideal circular shape. Based on the surrounding behavior score, it is determined whether the target is engaged in surrounding reconnaissance behavior within the current time window.
[0005] Preferably, the process of obtaining the cumulative rotation angle is as follows: Based on the position coordinates of the target and the coordinates of the center point of the protected facility at each time, the original interval angle of the target relative to the center point of the protected facility at each time is calculated using the inverse trigonometric function. Calculate the original interval angle difference between the current time and the adjacent previous time. If the original interval angle difference is less than -180°, then the cumulative rotation angle of the target at the current time is the sum of the cumulative rotation angle of the adjacent previous time, the original interval angle difference, and 360°. If the original interval angle difference is greater than 180°, then the target's cumulative rotation angle at the current moment is the sum of the cumulative rotation angle of the adjacent previous moment and the original interval angle difference minus 360°. If the absolute value of the original interval angle difference is less than or equal to 180°, then the cumulative rotation angle of the target at the current moment is the sum of the cumulative rotation angle of the adjacent previous moment and the original interval angle difference; where the initial cumulative rotation angle is the original interval angle when the target just enters the monitoring area.
[0006] Preferably, the radial constant distance is the average of the measured distances of the target at all times within the current time window.
[0007] Preferably, the process of constructing the distance matrix is as follows: Within the current time window, the measured distances of the target at all times are combined into a measured distance sequence, and a constant distance sequence with the same length as the measured distance matrix is constructed. In the constant distance sequence, all elements are radial constant distances. Construct a distance matrix where the element in the i-th row and j-th column represents the difference between the i-th measured distance in the measured distance sequence and the j-th radial constant distance in the constant distance sequence.
[0008] Preferably, the original shape deviation value of the target within the current time window is the sum of all elements on the optimal regular path in the distance matrix.
[0009] Preferably, the calculation process for the unidirectional rotation ratio is as follows: Calculate the sum of the cumulative rotation angle differences between all adjacent moments within the current time window, and denot it as the total rotation angle; Calculate the ratio of net rotation amplitude to the sum of rotation angles, and record it as the percentage of unidirectional rotation of the target within the current time window.
[0010] Preferably, the net rotation amplitude of the cumulative rotation angle within the current time window is: the absolute difference between the cumulative rotation angle of the target at the end of the preset time window and the cumulative rotation angle at the start of the window.
[0011] Preferably, the expression for the target's orbital behavior score within the current time window is: In the formula, This represents the target's orbital behavior score within the current time window; This indicates the percentage of unidirectional rotation of the target within the current time window; This represents the normalized value of the original shape deviation of the target within the current time window; This represents the preset shape sensitivity coefficient.
[0012] Preferably, the process of determining whether the target is conducting surround reconnaissance behavior within the current time window is as follows: If the target's orbiting behavior score is greater than the preset orbiting threshold within the current time window, it is determined that the target is engaging in orbiting reconnaissance behavior within the current time window; otherwise, it is determined that the target is not engaging in orbiting reconnaissance behavior within the current time window.
[0013] Secondly, embodiments of this application also provide a suspicious behavior identification system based on pedestrian trajectory and dwell analysis, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any of the above-described suspicious behavior identification methods based on pedestrian trajectory and dwell analysis.
[0014] This application has at least the following beneficial effects: This application constructs a dynamic radial constant distance and introduces a dynamic time warping algorithm to achieve robust quantitative analysis of the geometric shape of non-uniform motion trajectories. It effectively eliminates the interference of temporal redundancy such as pedestrian pauses and speed changes on shape matching, thereby accurately extracting the deviation features that reflect the true shape of the trajectory. This ensures that the original shape deviation value can objectively characterize the degree to which the target deviates from the ideal circle, greatly improving the accuracy and anti-interference ability to distinguish between circumventing reconnaissance and straight-line passing behavior in complex real-world scenarios. Furthermore, this application constructs a denominator-penalized scoring model that combines unidirectional rotation ratio with normalized morphological deviation, thereby achieving dual feature fusion of target motion direction consistency and trajectory geometry. This not only effectively eliminates interference from single features such as lingering in place or passing in a straight line, but also accurately maps the suspiciousness of the target through quantified circling behavior scores, which helps to improve the accuracy of identifying suspicious pedestrian behavior. Ultimately, this application achieves precise conversion from abstract quantitative scores to specific security alarm signals by setting a reasonable threshold judgment mechanism and continuous confirmation logic. It effectively balances recognition sensitivity and false alarm rate by using a preset surround threshold, and eliminates instantaneous noise interference through judgment conditions of multiple consecutive frames, ensuring that the alarm is triggered only when the target shows a stable and strong surround detection intention. This completes the automated identification and efficient early warning of suspicious behavior, and improves the accuracy of identifying suspicious pedestrian behavior. Attached Figure Description
[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0016] Figure 1 A flowchart illustrating the steps of a suspicious behavior identification method based on pedestrian trajectory and dwelling analysis provided in one embodiment of this application; Figure 2 This is a flowchart of the surrounding behavior determination process provided in one embodiment of this application. Detailed Implementation
[0017] To further illustrate the technical means and effects adopted by this application to achieve the intended inventive objective, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the suspicious behavior identification method and system based on pedestrian trajectory and dwelling analysis proposed in this application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.
[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.
[0019] The following description, in conjunction with the accompanying drawings, details the specific scheme of the suspicious behavior identification method and system based on pedestrian trajectory and dwelling analysis provided in this application.
[0020] Please see Figure 1 The diagram illustrates a flowchart of a suspicious behavior identification method based on pedestrian trajectory and dwelling analysis according to an embodiment of this application. The method includes the following steps: Step S1: Real-time acquisition of the measured distance and cumulative rotation angle from the target to the center point of the protection facility within the monitoring area.
[0021] S101: Reference calibration and initialization.
[0022] The geometric center of the protected facility (such as a guard post, vault door, or important equipment) can be marked on the monitoring screen through a human-computer interaction interface; that is, the center point of the protected facility. .
[0023] S102: Calculation of distance to monitoring point and original angle.
[0024] At time t, the position of the target in the image coordinate system is obtained through a multi-target tracking algorithm. To quantify the spatial relationship between the target and the protected facilities, the following calculations are performed: First, the target is calculated using the Euclidean distance formula. Center point of the protected facility The straight-line distance between them is taken as the measured distance from the target to the center point of the protected facility at time t. : Secondly, the original interval angle of the target relative to the center point of the protected facility at time t is calculated using inverse trigonometric functions. : The angle value output by this function is usually limited to... Within the interval, this means that when the target continuously moves in circles across the negative X-axis (e.g., from...). Turn to The values will undergo non-physical jumps.
[0025] S103: Angle jump elimination and cumulative rotation angle generation.
[0026] To analyze the continuous rotation trend of the target (e.g., to identify whether the target has completed one or two rotations), the aforementioned numerical jumps must be eliminated to generate a continuous, monotonic cumulative rotation angle. The specific real-time steps are as follows: Based on the position coordinates of the target and the coordinates of the center point of the protected facility at each time, the original interval angle of the target relative to the center point of the protected facility at each time is calculated using the inverse trigonometric function. Calculate the original interval angle difference between the current time and the adjacent previous time. If the original interval angle difference is less than -180°, then the cumulative rotation angle of the target at the current time is the sum of the cumulative rotation angle of the adjacent previous time, the original interval angle difference, and 360°. If the original interval angle difference is greater than 180°, then the target's cumulative rotation angle at the current moment is the sum of the cumulative rotation angle of the adjacent previous moment and the original interval angle difference minus 360°. If the absolute value of the original interval angle difference is less than or equal to 180°, then the cumulative rotation angle of the target at the current moment is the sum of the cumulative rotation angle of the adjacent previous moment and the original interval angle difference; where the initial cumulative rotation angle is the original interval angle when the target just enters the monitoring area.
[0027] At this point, the measured distance and cumulative rotation angle from the target to the center point of the protected facility within the monitoring area are obtained.
[0028] Step S2: Within the preset current time window, calculate the radial constant distance based on the measured distance from the target to the center point of the protected facility, which is used to characterize the radial characteristics of the ideal circular motion; construct a distance matrix between the measured distance and the radial constant distance, and use the dynamic time warping algorithm to search for the optimal warping path in the distance matrix to determine the original shape deviation value of the target within the current time window, which is used to characterize the geometric deviation of the measured distance relative to the radial constant distance of the ideal circular motion.
[0029] S201: Within the preset current time window, calculate the radial constant distance based on the measured distance from the target to the center point of the protected facility, which is used to characterize the radial features of the ideal circular motion.
[0030] To determine whether a target's trajectory conforms to the circumferential characteristic (i.e., the radial distance remains relatively constant), a dynamic reference baseline must first be constructed based on the current data. Since the distance between each target and the facility varies, a fixed radius value cannot be used; instead, a personalized template must be generated based on the target's own motion characteristics. Therefore, this embodiment calculates the constant radial distance based on the measured distance from the target to the center point of the protected facility, used to characterize the radial features of ideal circular motion. The specific process is as follows: In this embodiment, the average of the measured distances of the target at all times within the current time window is taken as the radial constant distance.
[0031] Based on this, in order to prevent numerical anomalies in subsequent calculations (such as the denominator approaching zero), anomaly boundary handling logic is introduced: it is determined whether the radial constant distance is less than the preset minimum safety radius (in this embodiment, the minimum safety radius is set to 0.5 meters). If the radial constant distance is less than this value, it is considered that the target has contacted or entered the facility, and a contact alarm is directly triggered and subsequent morphological analysis is terminated; otherwise, the subsequent steps are continued.
[0032] It should be noted that, to ensure the time window covers the minimum time span required for the target to complete its effective circling maneuver, the preset time window size should not be set to a fixed short duration (e.g., 2 seconds), but should be adapted to the spatial scale of the scene. Preferably, this length is set to the length of the perimeter of the protective facility when the target traverses at a normal walking speed (e.g., 1.0-1.5 m / s). to The number of time frames required to measure the radius in radians. For example, for a facility with a radius of approximately 5 meters, its... The perimeter is approximately 15 meters. If the target walking speed is 1.5 m / s, it will take approximately 10 seconds. At a sampling rate of 25 frames per second, it is recommended to set the preset time window to 250 frames. The end time of the preset time window is the current time.
[0033] S202: Construct a distance matrix between the measured distance and the radial constant distance, and use a dynamic time warping algorithm to search for the optimal warping path in the distance matrix to determine the original shape deviation value of the target within the current time window, which is used to characterize the geometric deviation of the measured distance relative to the ideal circular motion radial constant distance.
[0034] To accurately identify the target's movement intention, it is necessary to first extract feature indicators that reflect the trajectory geometry from the radial distance data. Therefore, this embodiment constructs a distance matrix between the measured distance and the constant radial distance, and uses a dynamic time warping algorithm to search for the optimal warping path in the distance matrix to determine the original shape deviation value of the target within the current time window. This quantifies the degree to which the target trajectory deviates radially from the "ideal circle," i.e., it determines whether the target is moving while maintaining a constant distance. The specific process is as follows: (1) Construct the distance matrix between the measured distance and the radial constant distance.
[0035] Within the current time window, the measured distances of the target at all times are combined into a measured distance sequence, and a constant distance sequence with the same length as the measured distance matrix is constructed. In the constant distance sequence, all elements are radial constant distances. Construct a distance matrix M, where the element in the i-th row and j-th column of the distance matrix M represents the difference between the i-th measured distance in the measured distance sequence and the j-th radial constant distance in the constant distance sequence.
[0036] (2) The optimal warping path is searched in the distance matrix using the dynamic time warping algorithm to determine the original shape deviation value of the target within the current time window, which is used to characterize the geometric deviation of the measured distance relative to the ideal circular motion radial constant distance.
[0037] Use dynamic programming to search for an optimal regular path from the bottom left vertex to the top right vertex in the distance matrix. This path search must satisfy the following constraints: a. Boundary constraints: The path must begin at the bottom left vertex of the distance matrix and end at the top right vertex; b. Continuity constraint: The path cannot skip steps; c. Monotonicity constraint: The path must move forward over time and cannot go backward.
[0038] The cumulative cost D is calculated using a recursive formula, where Indicates from the starting point to The minimum cumulative cost, This represents the value of the element in the i-th row and j-th column of the distance matrix M: Finally, the path corresponding to the minimum cumulative cost from the starting point to the ending point will be taken as the optimal regularization path, and the minimum cumulative cost from the starting point to the ending point, that is, the sum of all elements on the optimal regularization path, will be taken as the original morphological deviation value.
[0039] Based on the original shape deviation value, it can be understood that the original shape deviation value is used to characterize the geometric difference between the measured distance of the target and the constant radial distance of the ideal circular motion, reflecting the degree to which the radial distance of the target remains constant during the motion. Its calculation is mainly affected by the numerical difference between the measured distance and the constant radial distance and the degree of temporal alignment. When the measured distance deviates from the constant radial distance to a greater extent (such as the parabolic radial characteristics when passing through in a straight line), the original shape deviation value will increase significantly, reflecting that the target trajectory shape deviates seriously from the ideal circle, which has a strong inhibitory effect on the final orbiting behavior score, thus judging it as non-orbiting behavior. Conversely, when the measured distance always fluctuates closely around the constant radial distance (such as orbiting reconnaissance behavior), even if there are pauses or speed changes on the time axis, the original shape deviation value remains extremely low, reflecting that the target trajectory closely follows the ideal circle, which helps the system to identify the real orbiting behavior.
[0040] Thus, this embodiment achieves robust quantitative analysis of the geometric shape of non-uniform motion trajectories by constructing a dynamic radial constant distance and introducing a dynamic time warping algorithm. It effectively eliminates the interference of temporal redundancy such as pedestrian pauses and speed changes on shape matching, thereby accurately extracting the deviation features that reflect the true shape of the trajectory. This ensures that the original shape deviation value can objectively characterize the degree to which the target deviates from the ideal circle, greatly improving the accuracy and anti-interference ability to distinguish between circumventing reconnaissance and straight-line passing behavior in complex real-world scenarios.
[0041] Step S3: Based on the net rotation amplitude of the cumulative rotation angle within the current time window and the difference in cumulative rotation angle between all adjacent moments, calculate the unidirectional rotation ratio and combine it with the original shape deviation to generate the target's orbital behavior score within the current time window, which is used to characterize the comprehensive suspiciousness of the target that simultaneously satisfies the two characteristics of continuous unidirectional rotation and ideal circular shape.
[0042] A constant radial distance alone cannot fully confirm circling reconnaissance behavior, because the radial distance remains essentially constant even when the target is stationary or circling within a small area. Effective circling reconnaissance must be accompanied by continuous changes in the viewing angle. Therefore, this embodiment calculates the unidirectional rotation ratio based on the net rotation amplitude of the cumulative rotation angle within the current time window and the difference in cumulative rotation angles between all adjacent moments. This ratio, combined with the original morphological deviation, generates a circling behavior score for the target within the current time window. This score characterizes the comprehensive suspiciousness of a target that simultaneously satisfies both continuous unidirectional rotation and an ideal circular morphology. The specific process is as follows: First, this embodiment calculates the unidirectional rotation percentage based on the net rotation amplitude of the cumulative rotation angle within the current time window and the difference in cumulative rotation angle between all adjacent times. The specific process is as follows: In this embodiment, the absolute difference between the cumulative rotation angle of the target at the end of the preset time window and the cumulative rotation angle at the start time is taken as the net rotation amplitude of the cumulative rotation angle within the current time window.
[0043] Furthermore, the sum of the cumulative rotation angle differences between all adjacent moments within the current time window is calculated and denoted as the total rotation angle. In order to eliminate high-frequency noise interference caused by video detection frame jitter, an angle jitter dead zone threshold is set (2° in this embodiment). When the absolute value of the cumulative rotation angle difference between adjacent moments is less than the angle jitter dead zone threshold, the difference is set to 0 and is not included in the cumulative calculation.
[0044] The ratio of the net rotation amplitude to the sum of the rotation angles is recorded as the unidirectional rotation percentage of the target within the current time window. Before calculating the ratio, a preset parameter adjustment factor is added to the sum of the rotation angles to prevent the denominator from being 0. In this embodiment, the preset parameter adjustment factor is 0.01. In actual applications, as other implementation methods, implementers can also set it according to specific circumstances. This embodiment does not impose any special restrictions.
[0045] Based on the unidirectional rotation ratio, it can be understood that the unidirectional rotation ratio is used to characterize the consistency of the target's rotation direction within the current time window, reflecting whether the target is conducting continuous unidirectional circling motion. Its calculation is mainly affected by the net rotation amplitude and the sum of the rotation angles. When the net rotation amplitude is larger and closer to the sum of the rotation angles (i.e., the target is continuously rotating in one direction), the unidirectional rotation ratio approaches 1, reflecting that the target has extremely strong directional consistency, which is consistent with the rotation characteristics of circling reconnaissance and will significantly improve the circling behavior score. Conversely, when the target frequently wanders back and forth or swings in place, causing the net rotation amplitude to be much smaller than the sum of the rotation angles, the unidirectional rotation ratio approaches 0, reflecting that the target's rotation direction is chaotic and will significantly lower the circling behavior score, thereby effectively eliminating aimless wandering behavior.
[0046] Furthermore, this embodiment generates a target's orbital behavior score within the current time window based on the proportion of unidirectional rotation and the original morphological deviation. This score characterizes the comprehensive suspiciousness of a target that simultaneously satisfies both continuous unidirectional rotation and an ideal circular morphology. The specific process is as follows: As one implementation method, in this embodiment, the target's orbital behavior score within the current time window The expression is: In the formula, This indicates the percentage of unidirectional rotation of the target within the current time window; This represents the normalized value of the original shape deviation of the target within the current time window; This represents the preset shape sensitivity coefficient.
[0047] It should be noted that the preset shape sensitivity coefficient is set manually, and its value range is [3.0, 10.0]. In this embodiment, the preset shape sensitivity coefficient is set to 5.0. In actual applications, as other implementation methods, implementers can also set it according to specific circumstances. This embodiment does not impose any special restrictions.
[0048] It should be noted that the process of normalizing the original morphological deviation value in this embodiment is as follows: The expression for the normalized value of the original morphological deviation is: In the formula, This indicates the deviation value of the original shape; This represents the mean of all elements in a radial constant distance sequence; This represents a preset constant greater than 0, used to prevent the denominator from being 0. In this embodiment... The value of is 0.01. Provided that the denominator is not zero and does not excessively affect the calculation result, the implementer may also set it according to the specific situation. This embodiment does not impose any special restrictions.
[0049] Based on the orbiting behavior score, it can be understood that the orbiting behavior score is used to characterize the comprehensive suspiciousness of a target that simultaneously meets the two characteristics of continuous unidirectional rotation and ideal circular shape, reflecting the strength of the target's intention to conduct orbiting reconnaissance behavior. Its calculation is jointly affected by the proportion of unidirectional rotation (numerator) and the normalized original shape deviation value (denominator penalty term). When the proportion of unidirectional rotation is higher and the shape deviation value is smaller, the orbiting behavior score is higher, reflecting that the target maintains a constant radial distance while continuously rotating, which is highly likely to be orbiting reconnaissance behavior, and the system will trigger a high-level warning. Conversely, when the proportion of unidirectional rotation is low or the shape deviation value is large, the orbiting behavior score is significantly reduced, reflecting that the target is either rotating disorderly or the radial trajectory is divergent (such as passing in a straight line), which does not have the typical characteristics of orbiting reconnaissance, and will be judged as normal behavior to avoid false alarms.
[0050] Thus, this embodiment achieves the fusion of dual features of target motion direction consistency and trajectory geometry by constructing a denominator-penalized scoring model that combines unidirectional rotation ratio and normalized morphological deviation. This not only effectively eliminates interference from single features such as lingering in place or passing in a straight line, but also accurately maps the suspiciousness of the target through quantified circling behavior scores, which helps to improve the accuracy of identifying suspicious pedestrian behavior.
[0051] Step S4: Based on the surrounding behavior score, determine whether the target is engaged in surrounding reconnaissance behavior within the current time window.
[0052] The orbiting behavior score calculated through the above steps, by integrating the unidirectionality of rotation direction and the constancy of distance pattern, constructs a comprehensive index that can quantify the strength of a target's "orbiting reconnaissance intention." The orbiting behavior score effectively maps complex spatiotemporal trajectory characteristics into a single numerical value: a higher orbiting behavior score means that the target can maintain a good radial distance while continuously rotating, consistent with typical orbiting reconnaissance behavior patterns; a lower score means that the target's rotation direction is chaotic or its radial distance fluctuates wildly, tending more towards aimless wandering or straight-line movement. Based on the orbiting behavior score, the process of automatically identifying suspicious behavior to determine whether the target is engaging in orbiting reconnaissance within the current time window is as follows: In this embodiment, if the target's orbiting behavior score is greater than a preset orbiting threshold within the current time window, it is determined that the target has orbiting reconnaissance behavior within the current time window; otherwise, it is determined that the target does not have orbiting reconnaissance behavior within the current time window.
[0053] Preferably, the flowchart of the surrounding behavior determination process provided in this embodiment is as follows: Figure 2 As shown.
[0054] It should be noted that the preset circling threshold value of 0.8 in this embodiment is an empirical value calibration based on the distribution characteristics of unidirectional rotation ratio and morphological deviation features, aiming to achieve the optimal balance between recognition sensitivity and false alarm rate. Since the unidirectional rotation ratio of ideal circling behavior is close to 1 and the morphological deviation is close to 0, the circling behavior score is close to 1. Setting 0.8 as the threshold can effectively accommodate the score loss caused by non-ideal factors such as intermittent pauses of pedestrians and slight trajectory jitter in real-world scenarios. As long as the target maintains high rotation consistency and the radial distance is relatively stable, its score can exceed the threshold and be accurately identified, avoiding false alarms caused by an excessively high threshold. At the same time, this threshold is sufficient to suppress the scores of behaviors such as passing in a straight line (large morphological deviation leads to a sharp drop in score) or wandering in disorder (low rotation ratio leads to a decrease in score) below the threshold, thereby effectively preventing false alarms. In practical applications, as other implementation methods, implementers can also set their own thresholds according to specific circumstances, and this embodiment does not impose any special restrictions.
[0055] If the target's orbiting behavior score is greater than the preset orbiting threshold for the current time and the preceding and adjacent preset number of time windows, an alarm signal is immediately sent to the security management platform. The alarm signal includes the target's ID, current location coordinates, and the timestamp of the current time. The preset number is set manually. In this embodiment, the preset number is 4. In actual applications, as other implementation methods, the implementer can also set it according to the specific situation. This embodiment does not impose any special restrictions.
[0056] Thus, this embodiment achieves the accurate conversion from abstract quantitative scores to specific security alarm signals by setting a reasonable threshold judgment mechanism and continuous confirmation logic. It effectively balances recognition sensitivity and false alarm rate by using a preset surround threshold, and eliminates instantaneous noise interference by using judgment conditions for multiple consecutive frames. This ensures that the alarm is triggered only when the target shows a stable and strong surround detection intention, thereby completing the automated identification and efficient early warning of suspicious behavior and improving the accuracy of pedestrian suspicious behavior identification.
[0057] Based on the same inventive concept as the above methods, this application also provides a suspicious behavior identification system based on pedestrian trajectory and dwelling analysis, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above-described suspicious behavior identification methods based on pedestrian trajectory and dwelling analysis.
[0058] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments of this specification have been described above. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.
[0059] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.
[0060] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the principles of this application should be included within the protection scope of this application.
Claims
1. A method for identifying suspicious behavior based on pedestrian trajectory and dwell time analysis, characterized in that, The method includes the following steps: Real-time acquisition of the measured distance and cumulative rotation angle from the target to the center point of the protected facility within the monitoring area; Within the preset current time window, the radial constant distance is calculated based on the measured distance from the target to the center point of the protected facility, which is used to characterize the radial characteristics of the ideal circular motion; a distance matrix is constructed between the measured distance and the radial constant distance, and the optimal warping path is searched in the distance matrix using a dynamic time warping algorithm to determine the original shape deviation value of the target within the current time window, which is used to characterize the geometric deviation of the measured distance relative to the radial constant distance of the ideal circular motion; Based on the net rotation amplitude of the cumulative rotation angle within the current time window and the difference in cumulative rotation angle between all adjacent times, the unidirectional rotation ratio is calculated and combined with the original shape deviation to generate the target's orbital behavior score within the current time window, which is used to characterize the comprehensive suspiciousness of the target that simultaneously satisfies the two characteristics of continuous unidirectional rotation and ideal circular shape. Based on the surrounding behavior score, it is determined whether the target is engaged in surrounding reconnaissance behavior within the current time window.
2. The suspicious behavior identification method based on pedestrian trajectory and dwell time analysis as described in claim 1, characterized in that, The process of obtaining the cumulative rotation angle is as follows: Based on the position coordinates of the target and the coordinates of the center point of the protected facility at each time, the original interval angle of the target relative to the center point of the protected facility at each time is calculated using the inverse trigonometric function. Calculate the original interval angle difference between the current time and the adjacent previous time. If the original interval angle difference is less than -180°, then the cumulative rotation angle of the target at the current time is the sum of the cumulative rotation angle of the adjacent previous time, the original interval angle difference, and 360°. If the original interval angle difference is greater than 180°, then the target's cumulative rotation angle at the current moment is the sum of the cumulative rotation angle of the adjacent previous moment and the original interval angle difference minus 360°. If the absolute value of the original interval angle difference is less than or equal to 180°, then the cumulative rotation angle of the target at the current moment is the sum of the cumulative rotation angle of the adjacent previous moment and the original interval angle difference; where the initial cumulative rotation angle is the original interval angle when the target just enters the monitoring area.
3. The suspicious behavior identification method based on pedestrian trajectory and dwell time analysis as described in claim 1, characterized in that, The radial constant distance is the average of the measured distances of the target at all times within the current time window.
4. The suspicious behavior identification method based on pedestrian trajectory and dwell time analysis as described in claim 1, characterized in that, The process of constructing the distance matrix is as follows: Within the current time window, the measured distances of the target at all times are combined into a measured distance sequence, and a constant distance sequence with the same length as the measured distance matrix is constructed. In the constant distance sequence, all elements are radial constant distances. Construct a distance matrix where the element in the i-th row and j-th column represents the difference between the i-th measured distance in the measured distance sequence and the j-th radial constant distance in the constant distance sequence.
5. The suspicious behavior identification method based on pedestrian trajectory and dwell time analysis as described in claim 1, characterized in that, The original shape deviation of the target within the current time window is the sum of all elements on the optimal regular path in the distance matrix.
6. The suspicious behavior identification method based on pedestrian trajectory and dwell time analysis as described in claim 1, characterized in that, The calculation process for the unidirectional rotation ratio is as follows: Calculate the sum of the cumulative rotation angle differences between all adjacent moments within the current time window, and denot it as the total rotation angle; Calculate the ratio of net rotation amplitude to the sum of rotation angles, and record it as the percentage of unidirectional rotation of the target within the current time window.
7. The suspicious behavior identification method based on pedestrian trajectory and dwell time analysis as described in claim 6, characterized in that, The net rotation amplitude of the cumulative rotation angle within the current time window is: the absolute difference between the cumulative rotation angle of the target at the end of the preset time window and the cumulative rotation angle at the start time.
8. The suspicious behavior identification method based on pedestrian trajectory and dwell time analysis as described in claim 1, characterized in that, The expression for the target's orbital behavior score within the current time window is: In the formula, This represents the target's orbital behavior score within the current time window; This indicates the percentage of unidirectional rotation of the target within the current time window; This represents the normalized value of the original shape deviation of the target within the current time window; This represents the preset shape sensitivity coefficient.
9. The suspicious behavior identification method based on pedestrian trajectory and dwell time analysis as described in claim 1, characterized in that, The process for determining whether a target is conducting surround reconnaissance within the current time window is as follows: If the target's orbiting behavior score is greater than the preset orbiting threshold within the current time window, it is determined that the target is engaging in orbiting reconnaissance behavior within the current time window; otherwise, it is determined that the target is not engaging in orbiting reconnaissance behavior within the current time window.
10. A suspicious behavior identification system based on pedestrian trajectory and dwell analysis, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the suspicious behavior identification method based on pedestrian trajectory and dwelling analysis as described in any one of claims 1-9.