A method for crowd motion trend perception based on machine vision

Abstract

This invention discloses a crowd movement trend perception method based on machine vision. By detecting, tracking, and fusion analyzing multiple targets in a video sequence, it achieves stable perception of the overall movement direction and trend of the crowd, reduces the impact of individual abnormal behavior on the overall trend judgment, and can still output stable crowd movement direction and speed trend in complex scenarios such as occlusion, intersection, and dense crowds.

Description

Technical Field

[0001] This invention belongs to the field of computer vision and intelligent sensing technology, and relates to a technology for detecting, tracking, and analyzing the trajectory of multiple target groups based on machine vision, thereby realizing the perception and prediction of the overall movement trend of the crowd. Background Technology

[0002] With the development of smart devices and artificial intelligence technology, vision-based human detection and tracking technology has been widely used in fields such as security monitoring, smart transportation, business analysis, and service robots.

[0003] Existing technologies mainly focus on single-target tracking or individual behavior analysis, such as determining behavior by detecting the position, speed, or posture of a single pedestrian. However, in real-world applications, such as exhibition hall navigation, shopping malls, and transportation hubs, targets often present as groups of multiple people simultaneously, exhibiting the following characteristics: group dynamics: the movement of the crowd tends to show an overall trend rather than the behavior of independent individuals; dynamic nature: the direction and speed of the crowd continuously change over time; uncertainty: interference between individuals can easily lead to tracking drift or misjudgment; and complexity: occlusion, intersections, and clustering frequently occur.

[0004] The shortcomings of existing technologies include: the lack of a unified group behavior modeling mechanism for multi-target tracking results; the lack of effective methods for fusion analysis of multi-target trajectories; the difficulty in stably outputting the overall movement direction and trend of the crowd; and the susceptibility to interference from abnormal individual behaviors in complex scenarios, leading to unstable overall judgment.

[0005] Therefore, it is necessary to propose a method that can fuse and analyze multi-target visual information to stably perceive the overall movement trend of a crowd. Summary of the Invention

[0006] This invention provides a method for extracting the overall movement trend of a crowd from multi-target visual data; it provides an algorithm for fusing and modeling multi-target trajectories to improve the stability of trend judgment; it reduces the impact of individual abnormal behavior on the overall trend judgment; and it can still output stable crowd movement direction and speed trends in complex scenarios with occlusion, intersection, and dense crowds.

[0007] This invention proposes a machine vision-based method for perceiving crowd movement trends. By detecting, tracking, and performing trajectory fusion analysis on multiple targets in a video sequence, it achieves stable perception of the overall movement direction and trend of the crowd.

[0008] Step 1, data collection A sequence of consecutive video frames ,in The frame number is used to perform scale normalization, noise suppression, and foreground enhancement on the image, thereby improving the stability of subsequent detection and tracking.

[0009] Step two: Use a human detection algorithm to identify targets in each frame of the image to obtain a target set. Calculate the center point coordinates of each target bounding box. ,in Indicates time The detected first One goal, Indicates the first The goal is at any time x-coordinate Indicates the first The goal is at any time The ordinate.

[0010] Step 3: Employ a multi-target tracking method to associate targets across frames, assign a unique identifier to each target, and construct the time-series trajectories of all targets. ,in Indicates the first The first goal in The position of the frame.

[0011] Furthermore, if there is a short-term loss in the trajectory, interpolation compensation is performed to ensure the continuity of the trajectory.

[0012] Step 4: Calculate the velocity vector for each trajectory based on the changes in position between adjacent frames. and direction angle of motion ,in Indicates the first The goal is at any time The velocity vector, Indicates the time interval between adjacent video frames. Indicates the first The goal is at any time The motion direction angle is used to obtain the motion characteristics of each target.

[0013] Furthermore, calculate the first... The goal is at any time acceleration vector This helps in judging motion stability.

[0014] Step 5: Construct a weight function for each trajectory using a trajectory validity evaluation mechanism. ,but ,in Indicates the first The length of the trajectory, Indicates the first The velocity stability of the trajectory Indicates the first The detection confidence of the trajectory. Indicates the first The directional consistency of the trajectories, where L represents the longest trajectory among all trajectories. This represents the standard deviation of trajectory velocity fluctuation. Indicates the first The average direction angle of motion of the trajectory, This represents the average direction angle of motion for all trajectories. The weighting coefficients are used to give stable and reliable trajectories a higher weight in subsequent fusion, thereby reducing the impact of noisy trajectories and unstable targets on the overall trend judgment.

[0015] Step 6: Calculate the deviation between the direction angle of each trajectory and the mean. If the preset threshold is exceeded If an abnormal trajectory is detected, it will be considered an abnormal trajectory and removed to avoid individual abnormal behaviors from interfering with the overall trend.

[0016] Step 7: Weight and superimpose the velocity vectors of the remaining trajectories to obtain the composite motion vector of the group. Calculate the overall direction of motion The overall movement speed of the population was calculated using a weighted average. ,in Indicates the first The average velocity vector of the trajectory, , They represent the first The horizontal and vertical components of the trajectory velocity vector. Indicates the first The average velocity of each trajectory is used to perform group trend fusion calculation on the effective trajectory set.

[0017] Furthermore, calculate the trend consistency index. This indicates the degree to which the movement direction of each target is consistent with the overall trend, where... This indicates the number of valid trajectories, thereby assessing the stability and consistency of current population movements.

[0018] Furthermore, an exponential smoothing method is adopted. Alternatively, a sliding window averaging method can be used to smooth the trend results of consecutive frames, where represents Represents the exponential smoothing coefficient. Indicates time The direction angle of the group's movement is used to obtain a stable trend output result, thereby avoiding the impact of short-term fluctuations on trend judgment in the time dimension.

[0019] Step 8: Output the overall movement trend information of the crowd, including the direction angle of the group's movement. Average speed of group movement Consistency indicators The confidence scores and their corresponding scores are provided for use by upper-level systems.

[0020] The advantages of this invention compared to existing technologies include: achieving group-level behavior modeling, upgrading from single-target analysis to multi-target fusion analysis, and reflecting the behavioral characteristics of real crowds; improving the stability of trend judgment by effectively reducing noise interference through abnormal trajectory elimination and fusion calculation; adapting to complex scenarios, and still being able to stably output trends in densely populated and heavily obstructed environments; strong versatility, not dependent on specific application devices, and can be widely used in fields such as robotics, security, and transportation; and high computational efficiency, based on trajectory-level feature calculation, avoiding complex semantic analysis, and possessing real-time processing capabilities. Attached Figure Description

[0021] Figure 1 It is a schematic diagram of the trajectory.

[0022] Figure 2 This is a schematic diagram showing the fusion and output of crowd movement trends. Detailed Implementation

[0023] Taking an exhibition hall demonstration scenario as an example and referring to the accompanying drawings, this invention will be described in further detail. In an exhibition hall setting, visitors typically exhibit small-group movement characteristics, such as groups of 2 to 5 people. Visitors may move from one exhibition area to the next, with their trajectories showing a certain directional consistency in space. Some visitors may pause briefly, turn their heads, or move against the flow of traffic, interfering with the overall trend assessment. The movement trend of the crowd may experience short-term fluctuations; for example, visitors may pause briefly in front of an exhibit before moving on again.

[0024] Inside the exhibition hall, there is a guided tour device with display and interactive capabilities, equipped with a visual acquisition device to obtain information on visitor behavior.

[0025] The visual acquisition device continuously acquires video streams from the area in front of the exhibition hall, forming a continuous video frame sequence. A human detection model is used to identify all visitor targets in each frame of the image and extract their location coordinates. .

[0026] A multi-target tracking algorithm is used to associate targets across frames, assign a unique identifier to each visitor, and construct a time-series trajectory. ,like Figure 1 As shown, the velocity vector of each trajectory is calculated based on the changes in position between adjacent frames. and direction angle of motion .

[0027] Weights are assigned to each trajectory, taking into account factors such as trajectory length, velocity stability, and detection confidence. Visitors with longer trajectories and stable directions receive relatively higher weights, while targets that appear briefly or have unstable movements receive relatively lower weights.

[0028] Calculate the deviation between the direction of each trajectory and the overall median direction. If the deviation exceeds the preset threshold, it will be considered an abnormal trajectory and removed from the overall trend calculation.

[0029] The velocity vectors of the remaining trajectories are weighted and superimposed to obtain the composite motion vector of the crowd. Calculate the overall direction of motion The overall movement speed of the population was calculated using a weighted average. .

[0030] Calculate the consistency index If the value exceeds a preset threshold, the current population is considered to have a clear movement trend.

[0031] Exponential smoothing method The trend results within a continuous time window are smoothed to obtain a stable trend output.

[0032] Output crowd movement trend information, including overall movement direction Average speed Consistency indicators ,like Figure 2 As shown.

[0033] The upper-level system drives the guide equipment to move in the same direction and rhythm as the visitors based on trend information, thereby achieving a natural following experience.

Claims

1. A method for perceiving crowd movement trends based on machine vision, characterized in that, include: Step 1, data collection A sequence of consecutive video frames ,in Indicates the frame number, and performs scale normalization, noise suppression, and foreground enhancement on the image; Step two: Use a human detection algorithm to identify targets in each frame of the image to obtain a target set. Calculate the center point coordinates of each target bounding box. ,in Indicates time The detected first One goal, Indicates the first The goal is at a certain moment x-coordinate Indicates the first The goal is at a certain moment The ordinate; Step 3: Employ a multi-target tracking method to associate targets across frames, assign a unique identifier to each target, and construct the time-series trajectories of all targets. ,in Indicates the first The first goal in Frame position; Step 4: Calculate the velocity vector for each trajectory based on the changes in position between adjacent frames. and direction angle of motion ,in Indicates the first The goal is at any time The velocity vector, Indicates the time interval between adjacent video frames. Indicates the first The goal is at any time The motion direction angle is used to obtain the motion characteristics of each target; Step 5: Construct a weight function for each trajectory using a trajectory validity evaluation mechanism. ,but ,in Indicates the first The length of the trajectory, Indicates the first The velocity stability of the trajectory Indicates the first The detection confidence of the trajectory. Indicates the first The directional consistency of the trajectories, where L represents the longest trajectory among all trajectories. This represents the standard deviation of trajectory velocity fluctuation. Indicates the first The average direction angle of motion of the trajectory, This represents the average direction angle of motion for all trajectories. These are the weighting coefficients; Step 6: Calculate the deviation between the direction angle of each trajectory and the mean. If the preset threshold is exceeded If the trajectory is abnormal, it will be removed. Step 7: Weight and superimpose the velocity vectors of the remaining trajectories to obtain the composite motion vector of the group. Calculate the overall direction of motion The overall movement speed of the population was calculated using a weighted average. ,in Indicates the first The average velocity vector of the trajectory, , They represent the first The horizontal and vertical components of the trajectory velocity vector. Indicates the first The average velocity of the trajectory; Step 8: Output the overall movement trend information of the crowd, including the direction angle of the group's movement. Average speed of group movement Consistency Indicators And its corresponding confidence score.

2. The method for perceiving crowd movement trends based on machine vision according to claim 1, characterized in that, Step three includes: if there is a short-term loss in the trajectory, then interpolation compensation is performed.

3. The method for perceiving crowd movement trends based on machine vision according to claim 1, characterized in that, Step four includes: calculating the... The goal is at any time acceleration vector .

4. The method for perceiving crowd movement trends based on machine vision according to claim 1, characterized in that, Step seven includes: calculating the trend consistency index. This indicates the degree to which the movement direction of each target is consistent with the overall trend, where... This indicates the number of valid trajectories.

5. The method for perceiving crowd movement trends based on machine vision according to claim 1, characterized in that, Step seven includes: employing exponential smoothing. Alternatively, a sliding window averaging method can be used to smooth the trend results of consecutive frames, where represents Represents the exponential smoothing coefficient. Indicates time The direction angle of the group's movement is used to obtain a stable trend output result.

Images