Comprehensive identification method for pedestrian bridge crossing social force behavior preference

Abstract

The invention discloses a comprehensive recognition method for pedestrian bridge crossing social force behavior preferences, and relates to the field of pedestrian track recognition, and the method comprises the steps: analyzing the behavior preferences, including walking habits, preference paths and walking speeds, of pedestrians during bridge crossing through questionnaire survey; performing video monitoring on pedestrians on the bridge floor through field experiments to obtain pedestrian bridge crossing video data under different working conditions; carrying out detection and trajectory tracking on the pedestrian bridge crossing video data based on a YOLOv11x target detection model and a DeepSORT tracking algorithm, and obtaining a pedestrian position and a motion trajectory; through questionnaire survey data and field experiment data, optimizing parameters of the trajectory model when the cellular automaton pedestrian passes the bridge; and applying the optimized cellular automaton pedestrian bridge crossing track model to comprehensive identification of pedestrian bridge crossing social force behavior preferences. The method solves the problem that a traditional method is difficult to accurately capture and describe the actual walking track of the pedestrian.

Description

Technical Field

[0001] The present invention relates to the field of pedestrian trajectory recognition, and in particular to a comprehensive recognition method for the social force behavior preference of pedestrians crossing a bridge. Background Art

[0002] With the acceleration of urbanization and the continuous rise in urban population density, traffic congestion has become increasingly prominent. As key nodes in urban transportation networks, overpasses shoulder the important mission of ensuring pedestrian safety and alleviating traffic pressure. However, their safety and comfort face numerous challenges, especially the pedestrian trajectory on the bridge deck, which is crucial for bridge seismic design, improving bridge comfort, and optimizing traffic flow.

[0003] Previous studies have often simplified pedestrian loads as constant loads, ignoring key factors such as pedestrian rigidity, mass, and damping. This has led to significant discrepancies between calculated results and actual conditions. Furthermore, traffic conditions on pedestrian bridges are far more complex than on motorway roads, and pedestrian paths are highly random, making it difficult for traditional research methods to accurately capture and describe pedestrians' actual trajectories. Summary of the Invention

[0004] In response to the above-mentioned deficiencies in the prior art, the present invention provides a comprehensive identification method for the social force behavior preference of pedestrians crossing a bridge, which solves the problem that traditional methods are difficult to accurately capture and describe the actual walking trajectory of pedestrians.

[0005] In order to achieve the above-mentioned object, the technical solution adopted by the present invention is: a comprehensive identification method for the social force behavior preference of pedestrians crossing a bridge, comprising the following steps:

[0006] S1: Analyze pedestrians' behavioral preferences when crossing bridges through questionnaire surveys, including walking habits, preferred paths, and walking speeds;

[0007] S2: Conduct video surveillance of pedestrians on the bridge deck through field experiments to obtain video data of pedestrians crossing the bridge under different working conditions;

[0008] S3: Detect and track pedestrians crossing the bridge video data based on the YOLOv11x target detection model and DeepSORT tracking algorithm to obtain the pedestrian's position and movement trajectory;

[0009] S4: Optimize the parameters of the cellular automaton pedestrian trajectory model when crossing the bridge based on questionnaire survey data and field experiment data;

[0010] S5: The optimized cellular automaton pedestrian trajectory model when crossing a bridge is used for comprehensive identification of the social force behavior preferences of pedestrians crossing a bridge.

[0011] Furthermore, the questionnaire survey in S1 includes two stages: preliminary survey and formal survey;

[0012] The preliminary survey verified the reliability of the questionnaire through reliability analysis and validity analysis;

[0013] The formal investigation was based on a multi-scenario setting, including static obstacles, dynamic obstacles, emergencies, and high-density crowd flow scenarios.

[0014] Furthermore, the reliability analysis was tested by Cronbach's α coefficient;

[0015] The validity analysis was tested by exploratory factor analysis.

[0016] Furthermore, the working conditions in S2 include:

[0017] a: Under the condition of an unobstructed bridge surface, select 5-15 people to cross the bridge surface, with a male-to-female ratio of 2:1 and an age range of 19-25 years old;

[0018] b: When there are obstacles on the bridge deck, 5-15 people are selected, with a male-to-female ratio of 2:1 and an age range of 19-25 years old.

[0019] Furthermore, in S3, the position information of pedestrians is extracted from the video frames through the YOLOv11x target detection model; the position changes of pedestrians between consecutive frames are obtained through the DeepSORT tracking algorithm, thereby constructing the pedestrian motion trajectory.

[0020] The beneficial effects of the present invention are:

[0021] The present invention comprehensively identifies the behavioral preferences of pedestrians crossing a bridge through video experiments and questionnaire analysis. The optimized prediction model can more accurately predict the walking trajectory of pedestrians when crossing a bridge.

[0022] The present invention sets obstacles on the bridge deck to simulate the walking behavior of pedestrians in different scenarios. Through a comprehensive recognition method, the behavioral preferences of pedestrians are obtained when crossing the bridge in different scenarios, considering the influence of social forces.

[0023] This paper uses the YOLOv11x target detection model and the DeepSORT tracking algorithm to achieve automatic and accurate detection and analysis of pedestrian trajectories.

[0024] The present invention comprehensively identifies the behavioral preferences of pedestrians when crossing a bridge in different scenarios, optimizes the parameters of a cellular automaton pedestrian trajectory model when crossing a bridge, and improves the accuracy and generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a flow chart of a comprehensive identification method for social force behavior preferences of pedestrians crossing bridges according to the present invention.

[0026] Figure 2This is the video analysis result of the scene without roadblocks.

[0027] Figure 3 This is the video analysis result of the roadblock scene.

[0028] Figure 4 This is a Matplotlib precision graph of a scene without roadblocks.

[0029] Figure 5 Matplotlib precision graph for a roadblock scenario.

[0030] Figure 6 This is a trajectory diagram of five people.

[0031] Figure 7 This is a trajectory diagram of ten people.

[0032] Figure 8 This is a graph showing the change in the mean trajectories of the five people.

[0033] Figure 9 This is a graph showing the changes in the mean trajectories of ten people. DETAILED DESCRIPTION

[0034] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0035] like Figure 1 As shown, a comprehensive identification method for the social force behavior preference of pedestrians crossing a bridge includes the following steps:

[0036] S1: Analyze pedestrians' behavioral preferences when crossing bridges through questionnaire surveys, including walking habits, preferred paths, and walking speeds;

[0037] S2: Conduct video surveillance of pedestrians on the bridge deck through field experiments to obtain video data of pedestrians crossing the bridge under different working conditions;

[0038] S3: Detect and track pedestrians crossing the bridge video data based on the YOLOv11x target detection model and DeepSORT tracking algorithm to obtain the pedestrian's position and movement trajectory;

[0039] S4: Optimize the parameters of the cellular automaton pedestrian trajectory model when crossing the bridge based on questionnaire survey data and field experiment data;

[0040] S5: The optimized cellular automaton pedestrian trajectory model when crossing a bridge is used for comprehensive identification of the social force behavior preferences of pedestrians crossing a bridge.

[0041] The questionnaire survey in S1 includes two stages: preliminary survey and formal survey;

[0042] The preliminary survey verified the reliability of the questionnaire through reliability analysis and validity analysis;

[0043] The reliability analysis was tested by Cronbach's α coefficient;

[0044] Cronbach's alpha is the most commonly used statistical method for assessing the consistency between items in a scale or questionnaire. It is calculated based on the variance of the scores on each item and the variance of the total score, thereby assessing the internal consistency of the entire scale. Cronbach's alpha values range from 0 to 1, with higher values indicating better internal consistency. Generally, a Cronbach's alpha greater than 0.6 is considered an acceptable level of reliability, 0.7 and above indicate good reliability, and values greater than 0.9 may indicate that the measurement instrument is overly simple or has other problems.

[0045] The alpha coefficient of the questionnaire for the study on bypassing pedestrian bridge obstacles is 0.680, and the alpha coefficients of each variable are shown in Table 1. The alpha coefficient of the entire questionnaire is 0.680, which is acceptable if it is greater than 0.6. The alpha coefficients of each variable are all greater than 0.6, which still meet the requirements of the scale reliability test.

[0046] Table 1 α coefficients of various variables

[0047]

[0048] The validity analysis was tested by exploratory factor analysis.

[0049] The second step in questionnaire analysis is to test the reliability of the scale. Reliability and validity are not equivalent concepts. High reliability does not necessarily mean high validity, but low reliability also does not guarantee high validity. Reliability examines the consistency of all items in a scale, while validity examines the effectiveness of each item—that is, whether each item contributes significantly to the scale. There are two statistical methods for testing validity: Exploratory Factor Analysis (EFA), using SPSS software, and Confirmatory Factor Analysis (CFA), using AMOS software. For scales with known dimensionality or established data, weak Confirmatory Factor Analysis (CFA) is necessary to verify the accuracy of the known dimensionality. For scales with unknown dimensionality, Exploratory Factor Analysis (EFA) is used to test validity. This not only examines the validity of each item but also scientifically explores the scale's dimensionality. Since the scale's dimensionality is unknown during the questionnaire pre-survey phase, Exploratory Factor Analysis was used to test the scale's validity.

[0050] Before implementing factor analysis on survey data, the first step is to conduct a suitability assessment to confirm whether the data meets the prerequisites for factor analysis. This assessment usually includes the KMO (Kaiser-Meyer-Olkin) value test and Bartlett's Test of Sphericity. The KMO value of the questionnaire data reached 0.676 (as shown in Table 2), which exceeds the threshold of 0.6, indicating that the data is suitable for factor analysis. At the same time, the results of the Bartlett's Test of Sphericity show that its significance probability is lower than the significance level of 0.01, further confirming that the data is suitable for factor analysis. Combining the results of the above two tests, it can be determined that the questionnaire data meets the prerequisites for factor analysis.

[0051] Table 2 KMO and Bartlett test

[0052]

[0053] Using SPSS software to conduct factor analysis on the questionnaire data, we analyzed the total variance explained by the 13 items in the questionnaire (as shown in Table 3) and divided them into four dimensions. The cumulative contribution of these four dimensions reached 61.437%, exceeding the generally accepted threshold of 60%. Therefore, this four-dimensional division can be considered relatively reasonable.

[0054] Table 3 Total variance explained

[0055]

[0056]

[0057] Subsequently, the component matrix of each item (shown in Table 4) was further examined to verify the validity of each item. Four components were extracted using principal component analysis. Item T14's factor loadings on both dimensions exceeded 0.5, indicating that it failed the validity test and was therefore deemed invalid and deleted. On the other hand, items T3 and T10 did not exhibit significant factor loadings on multiple dimensions, likely due to low correlations between the relevant variables in the dataset. It is recommended that these two items be considered separately in conjunction with specific research questions.

[0058] Table 4 Component matrix

[0059]

[0060] In addition, the remaining items all showed factor loadings higher than 0.5 on only a single dimension, indicating that they passed the validity test and were therefore retained in the questionnaire. In summary, based on the results of the above factor analysis, it can be concluded that the pre-survey questionnaire met the requirements in terms of validity and reliability, and therefore the results of the questionnaire pre-survey are valid.

[0061] The formal investigation was based on a multi-scenario setting, including static obstacles, dynamic obstacles, emergencies, and high-density crowd flow scenarios.

[0062] There were 119 participants in this questionnaire, mainly students aged 18-30, and their relatives and friends.

[0063] Scenario 1 involves a static obstacle on the bridge. When a pedestrian bridge has a roadblock in the middle, 60.5% of people, influenced by Chinese traffic habits, will choose to detour to the right, 28.57% will choose to detour to the left, and 10.92% will choose to walk directly across. If there are multiple roadblocks on the bridge, 88.43% of pedestrians will slow down, 8.26% will keep their speed constant, and 3.31% will speed up.

[0064] When an obstacle on a bridge is small enough for pedestrians to easily cross, 60.5% of pedestrians choose to cross directly, while 36.97% choose to detour. This shows that pedestrians consider both speed and safety. If there are clear signs next to the obstacle indicating the direction to avoid, only slightly more pedestrians will follow the instructions (54.55%) than those who will independently determine the detour direction based on the actual situation (45.45%).

[0065] From this, we can see that pedestrians will follow the instructions on the signs, but they will also make judgments based on their own situation and the actual situation when passing. This shows that pedestrians can both follow instructions and have the ability to make judgments. However, considering that there will be individual errors in judgment, the subjective consciousness of pedestrians should be taken into consideration when setting up signs.

[0066] In summary, when faced with static obstacles, pedestrians will cross the overpass on the right side in a quick, safe, and convenient manner. Lizhong Yang et al. studied the impact of pedestrians' rightward shift preference on critical density and provided a CA model to simulate pedestrian counterflow in a channel. k = 8 was proven to be reasonable under the considered conditions. People prefer to walk on the right side of the road or channel. This asymmetric behavior should not be ignored when studying flow rate or pattern formation. Signs have a certain guiding effect on pedestrians, but many people still rely on their own judgment.

[0067] Scenario 2 involves a bridge with dynamic obstacles, such as pedestrians or machinery, and a congested surface. In these situations, pedestrians either followed the crowd (47.11%) or followed their own judgment (49.59%). Only 3.31% waited until the obstacle was completely removed before proceeding, indicating that pedestrians prioritized speed and efficiency. This may be due to the fact that most of the respondents were young. Safety is also a key factor, so it is crucial to provide appropriate signage on crowded overpasses to ensure safe and efficient passage.

[0068] Scenario 3 involves an emergency such as a safety incident on an overpass. 34.45% of people choose to immediately run to the side of the bridge, seeking the nearest exit or shelter. 40.34% try to remain calm, observe their surroundings, and slowly evacuate along the path that appears safest. These two decisions are the most common. Another 12.4% choose to follow the crowd, demonstrating that people tend to trust their own judgment more in this situation. This suggests that emergency escape signs can be appropriately placed on overpasses to guide pedestrians to the correct escape routes. A study of an evacuation model based on cellular automata shows that panic has a significant impact on pedestrian movement. Increased panic levels prompt pedestrians to prioritize escaping danger, which often leads to passive evacuation to exits, potentially causing stampedes and prolonging evacuation time.

[0069] In scenario 4, when a large-scale event is set, the number of pedestrians on the overpass increases significantly. 88.43% of people choose to detour to quickly pass through, motivated by two factors: speed and safety. Only 11.57% choose to stick to their original planned route. This shows that route changes are closely related to group behavior. For example, if pedestrians are very close to an exit, they are less likely to change their route. Route changes tend to decrease over time. When the probability of route change is low (K < 0.05), evacuation times tend to be higher. However, when the probability of route change increases (K > 0.05), these safety measures tend to remain unchanged. This result may indicate that in environments where fewer route changes are expected, evacuation delays are inevitable. This is most pronounced in high-density environments. This behavioral pattern can be applied to evacuations during large-scale events, such as staggering arrival and departure times and adding temporary passages.

[0070] Combining the analysis of the above scenarios and the ranking of subjective factors in the questionnaire, we can conclude that pedestrians prioritize speed and smooth passage when crossing an overpass, followed by safety and convenience, and finally, whether their schedule is urgent. While herd mentality may influence pedestrians' habits when crossing overpasses, it is not the dominant factor.

[0071] For the field experiment, the South Gate Overpass at Beijing Jiaotong University was chosen as the appropriate location for this example, as the slope variation is negligible and would not significantly affect the experimental results. The experiment took place from September 27th to December 6th. The South Gate Overpass features a steel box girder structure, a smooth asphalt pavement free of wear and potholes, mature trees on both sides, and normal vehicle traffic under the bridge, meeting the characteristics of a typical pedestrian overpass. Therefore, this location was suitable for the study. The South Gate Overpass has a span of 17.6m, a width of 4m, and a height of 4.9m.

[0072] During the experiment, a camera mounted on a 2.8-meter-high plastic tube recorded video, ensuring that pedestrians were unaware of the camera's presence to avoid disrupting their behavior. The experiment was conducted under two conditions: the first involved 5-15 people crossing the bridge with no obstacles, with a male-to-female ratio of 2:1 and ages 19-25. The second involved 5-15 people crossing the bridge with obstacles installed to study pedestrian preferences when faced with obstacles, also with a male-to-female ratio of 2:1.

[0073] The camera used was a TP-LINK model with 4 megapixels, a 12mm focal length, a resolution of 2560 x 1440 pixels at 25 fps, and a lens distortion rate of 0.543%. The camera was 8.8 meters and 3.5 meters from the center of the bridge deck, respectively. Video analysis used Yolo8 to extract the position of each participant, with a maximum error of approximately 8 centimeters. The extracted point was the midpoint between each participant's feet. The video had 25 frames per second, and the participant's position was obtained every ten frames.

[0074] The working conditions in S2 include:

[0075] a: Under the condition of an unobstructed bridge surface, select 5-15 people to cross the bridge surface, with a male-to-female ratio of 2:1 and an age range of 19-25 years old;

[0076] b: When there are obstacles on the bridge deck, 5-15 people are selected, with a male-to-female ratio of 2:1 and an age range of 19-25 years old.

[0077] For this experiment, we organized 20 volunteer pedestrians to conduct the experiment on the Jiaotong University Road overpass. The bridge is approximately 17.6 meters long and 4 meters wide. We designed various scenarios with varying numbers of people and road conditions, and conducted multiple experiments. After setting up a camera on the south side of the bridge, we recorded the entire process and conducted the following four experiments.

[0078] Working condition 1: Bridge deck is barrier-free

[0079] This working condition is designed to simulate the walking habits of pedestrians when the bridge deck is clear.

[0080] Specific arrangements: Volunteers were divided into four groups with increasing numbers of 5, 10, 15, and 20 people to simulate road conditions with different levels of congestion, and then experiments were conducted to obtain pedestrians' walking habits. Each group walked back and forth on the bridge in the same direction twice to obtain sufficiently rich video footage.

[0081] Working condition 2: There is a narrow and long obstacle on the bridge deck

[0082] This working condition is designed to simulate the walking habits of pedestrians when the bridge deck is crowded.

[0083] Specific arrangements: Place 4 roadblocks on the center line of the bridge (the following obstacles are all simulated by roadblocks), randomly select 15 volunteers, and let the volunteers walk freely back and forth on the bridge twice under such road conditions. After obtaining the video, analyze the walking habits of pedestrians when it is more crowded.

[0084] Working condition 3: There is a large obstacle on the bridge deck

[0085] This working condition is designed to explore the walking habits of pedestrians when the bridge deck is very crowded.

[0086] Specific arrangements: Arrange the roadblocks into a larger diamond shape, leaving open paths on both sides wide enough for at most two people to pass through, and randomly select 15 people to walk back and forth twice under such road conditions to obtain the walking habits of pedestrians when the road is very crowded.

[0087] Working condition 4: There are many scattered obstacles on the bridge deck

[0088] This working condition is designed to study the walking habits of pedestrians when the road conditions on the bridge deck are more complicated.

[0089] Specific arrangement: Roadblocks are randomly placed on the bridge deck, and 15 volunteers are randomly selected to walk back and forth twice under such road conditions to obtain the walking habits of pedestrians under cluttered obstacle conditions.

[0090] After the experiment, the video captured during the experiment was processed with digital images and compared with the conclusions of the questionnaire analysis.

[0091] In S3, the YOLOv11x target detection model is used to extract the position information of pedestrians from the video frames; the DeepSORT tracking algorithm is used to obtain the position changes of pedestrians between consecutive frames, thereby constructing the pedestrian motion trajectory.

[0092] This study validated and optimized the survey results through video detection experiments. In the results analysis, relevant clips were extracted from the experimental videos for in-depth analysis, targeting specific scenarios in the questionnaire. Based on this principle, the experimental videos were first compressed and cropped, followed by pedestrian trajectory recognition.

[0093] This experiment uses the YOLOv11x object detection model combined with the DeepSORT tracking algorithm to analyze the movement trajectories of pedestrians on a pedestrian bridge. The YOLO family of models holds a prominent position in the field of object detection due to its high efficiency and real-time performance. The experiment is divided into two phases: object detection and trajectory tracking.

[0094] (1) Target detection stage:

[0095] Use the YOLOv11x model to detect pedestrians in video frames. YOLOv11x is a real-time object detection algorithm that can efficiently and accurately identify a variety of objects, including pedestrians, in a variety of scenarios. This model can be used to extract the location information (bounding box) of pedestrians from video frames.

[0096] (2) Trajectory tracking stage:

[0097] Detected pedestrians are tracked using the DeepSORT algorithm. DeepSORT combines deep learning features with the Hungarian algorithm to achieve robust trajectory tracking in complex scenes. This algorithm captures the position change of each pedestrian between consecutive frames, thereby constructing their motion trajectory.

[0098] The core of the experiment is to achieve automated and accurate detection and analysis of pedestrian trajectories on the footbridge through the combination of object detection and trajectory tracking. Furthermore, the experiment considers the impact of varying pedestrian density and roadblock layouts on pedestrian trajectories. By comparing trajectory data under different conditions, an in-depth analysis of their impact on pedestrian mobility is conducted.

[0099] The video dataset used in this study contains seven different pedestrian crossing bridge scenes. The specific working conditions are as follows:

[0100] (1) No roadblock scenario: 10 people one-way traffic; 15 people one-way traffic; 20 people one-way traffic; 25 people one-way traffic.

[0101] (2) Scenarios with roadblocks: one-way pedestrian traffic under square roadblock conditions; one-way pedestrian traffic under diamond roadblock conditions; one-way pedestrian traffic under straight roadblock conditions.

[0102] Each video records the flow of pedestrians on a pedestrian bridge. By processing and analyzing this video data, we can gain a deeper understanding of pedestrian trajectory characteristics and behavior patterns under different conditions, providing data support for the questionnaire model.

[0103] The video analysis results of seven working conditions are as follows: Figure 2 and Figure 3 As shown. Matplotlib drawing library is used to draw trajectory diagrams of pedestrian trajectory videos under seven working conditions. The trajectory diagrams of various working conditions are as follows Figure 4 and Figure 5shown.

[0104] In this study, both video analysis and meta-cellular automata models were used to study pedestrian trajectories and behavioral preferences. By comparing these two methods, we can better understand their advantages and limitations in pedestrian behavior simulation.

[0105] Based on real-world video data, pedestrian positions and trajectories are extracted using object detection (such as YOLO11x) and trajectory tracking (such as DeepSORT). This approach has the advantage of directly reflecting pedestrian behavior in real scenes. The data is derived from actual observations, resulting in high authenticity and reliability. However, it relies on video quality and the accuracy of the detection algorithm, and processing large amounts of video data requires significant computing resources.

[0106] Computational models based on discrete time and space simulate the dynamic behavior of complex systems by defining simple local rules. In pedestrian behavior research, each "cell" can represent a pedestrian or a spatial unit, and pedestrian behavior is determined by local rules (such as movement rules and avoidance rules). The advantages are simple models, high computational efficiency, and the ability to quickly simulate pedestrian behavior in different scenarios. However, the accuracy of the model depends on the setting of local rules and may not fully reflect the complex behavior in real-world scenarios.

[0107] By analyzing the experimental videos and questionnaire results, and comparing the outputs of the video analysis and meta-cellular automaton model, we found the following similarities and differences: Consistency lies in the high degree of agreement between the video analysis and meta-cellular automaton model results regarding pedestrian walking preferences (such as walking on the right and herd mentality). For example, 67% of pedestrians in the experimental videos walked on the right, which generally aligns with the preferences of pedestrians in the questionnaire survey. However, differences exist in high-density scenarios (such as one-way traffic with 20 or 25 people). Video analysis can more accurately capture pedestrian trajectory changes, while the meta-cellular automaton model may not fully reflect the real situation due to the limitations of its rule settings. In complex scenarios (such as obstacles on a bridge), the meta-cellular automaton model requires more complex rules to simulate pedestrian avoidance behavior, while video analysis can directly extract these behavioral characteristics.

[0108] By analyzing and processing multiple pedestrian trajectory data, individual differences and random noise can be eliminated, representative walking patterns can be extracted, and more representative data support can be provided for pedestrian behavior analysis, thereby revealing the commonalities and laws of pedestrian behavior.

[0109] For each independent variable point, calculate the average value of the dependent variable at that point for all curves. The formula is:

[0110]

[0111] in, is the value of the mean curve at x, N is the number of curves, y i (x) is the value of the i-th curve at position x.

[0112] The trajectory diagrams of five people and ten people are as follows: Figure 6 and Figure 7 As shown in Figure 2, the mean path of pedestrians predicted by the cellular automaton in a natural obstacle-free state is calculated using Matlab, as shown in Figure 2. Figure 8 and Figure 9 As shown:

[0113] Dynamic Time Warping (DTW) is a method used to measure the similarity between two time series. The core of DTW is to find an optimal warping path between the points of the two time series, so that the sum of the distances between the two series along this path is minimized.

[0114] When constructing a path, we not only consider the distance between individual points, but also calculate the cumulative distance from the starting point of the sequence to the current point. Suppose there are two time series X = [x1, x2, ..., x m ],Y=[y1,y2,…,y n ]. Let D(i,j) represent the cumulative distance between the 1st to ith point of sequence X and the 1st to jth point of sequence Y. Then the calculation of D(i,j) is based on the distance d(x i ,y i ) and the cumulative distances of its adjacent points are generally calculated using dynamic programming methods, i.e.

[0115]

[0116] After calculating all possible cumulative distances, we start from the last point D(m,n) and backtrack to find the optimal path. During backtracking, according to the cumulative distance calculation rules, we select the adjacent point that minimizes the cumulative distance as the previous point on the path each time until we reach the starting point. This results in a regular path that makes the two time series most similar. The cumulative distance on this path is the DTW distance between the two time series, which is used to measure their similarity.

[0117] According to the calculation, the obtained curve similarity is shown in the following table:

[0118] Table 5 DTW curve similarity

[0119]

[0120] The DTW distances are 100.1299 and 103.4472, respectively. These relatively large values indicate that the two lines differ significantly in shape and trend. As mentioned earlier, the larger the DTW distance, the lower the similarity between the time series, making it difficult to achieve a good match between these two lines through time warping.

[0121] The causes of errors are:

[0122] (1) Video detection relies on the actual captured video footage, and its data quality is affected by factors such as the performance of the shooting equipment, shooting angle, and lighting conditions. For example, a low-resolution camera may not be able to clearly capture the complete path information of the target, resulting in inaccurate detection results.

[0123] (2) Video detection requires preprocessing operations such as image filtering, noise reduction, and target extraction. If not handled properly, errors may be introduced.

[0124] (3) The data basis of cellular automata prediction is an abstract modeling of real scenes, which is usually generated according to preset rules and initial conditions and is essentially different from the real data in the actual video.

[0125] High-precision trajectory data obtained from video analysis is used to optimize the local rules of the meta-cellular automaton model. For example, pedestrian avoidance rules and speed change rules are adjusted based on video analysis results. In experiments, when an obstacle appeared on the bridge deck, video analysis showed that 65% of pedestrians chose to go around the obstacle from the right. This data can be used to adjust the avoidance rules in the meta-cellular automaton model to more closely resemble real-world behavior.

[0126] Introduce more pedestrian behavior characteristics into the meta-cellular automaton model, such as the degree of panic and the impact of individual differences (such as age and gender) on behavior. For example, in emergency scenarios, the model can introduce a panic factor to simulate pedestrians' rapid escape behavior. According to questionnaire survey results, pedestrians will combine signage and their own judgment to choose a detour when facing obstacles. This behavior can be used as a new rule in the meta-cellular automaton model to improve the model's accuracy and adaptability. Dynamically adjust the rules of the meta-cellular automaton based on dynamic changes in the scene (such as emergencies, the appearance of obstacles, etc.) to enable it to better simulate pedestrian behavior in complex scenarios.

[0127] By optimizing the cellular automaton model and video analysis, the performance of both methods in pedestrian behavior simulation has been significantly improved. The following is a comparison of the optimized models:

[0128] (1) Trajectory Accuracy: In the absence of roadblocks, the optimized cellular automaton model can more accurately simulate pedestrian trajectories, reducing the deviation from the video analysis results from the expected 10% before optimization to within 5%. In the presence of roadblocks, the consistency of the optimized model with the video analysis results in simulating pedestrian avoidance behavior increased from the expected 60% before optimization to approximately 80%.

[0129] (2) Behavioral characteristic simulation: In emergency scenarios, the optimized meta-cellular automaton model can better reflect pedestrian panic behavior and herd mentality. For example, the panic factor introduced in the model can make the simulated pedestrian's rapid escape behavior match the video analysis results to an expected degree of 75%. In complex scenarios (such as bridges with multiple obstacles), the optimized model can more accurately simulate the dynamic avoidance behavior of pedestrians, and the deviation from the video analysis results is reduced from the expected 20% before optimization to about 10%.

[0130] (3) Computational efficiency: Video analysis still requires high computing resources when processing large amounts of data, but by introducing multi-view fusion and optimizing detection algorithms, its processing efficiency is expected to increase by 30%. After optimization, the meta-cellular automaton model can more efficiently simulate the behavior of large groups of people, with an expected 50% increase in computational efficiency and the ability to quickly adapt to changes in different scenarios.

[0131] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and it should be understood that the scope of protection of the present invention is not limited to such specific descriptions and embodiments. Those skilled in the art can make various other specific variations and combinations based on the technical teachings disclosed in the present invention without departing from the essence of the present invention, and such variations and combinations are still within the scope of protection of the invention.

Claims

1. A comprehensive identification method for the social force behavior preference of pedestrians crossing bridges, characterized by: The following steps are involved: S1: Analyze pedestrians' behavioral preferences when crossing bridges through questionnaire surveys, including walking habits, preferred paths, and walking speeds; S2: Conduct video surveillance of pedestrians on the bridge deck through field experiments to obtain video data of pedestrians crossing the bridge under different working conditions; S3: Detect and track pedestrians crossing the bridge video data based on the YOLOv11x target detection model and DeepSORT tracking algorithm to obtain the pedestrian's position and movement trajectory; S4: Optimize the parameters of the cellular automaton pedestrian trajectory model when crossing the bridge based on questionnaire survey data and field experiment data; S5: The optimized cellular automaton pedestrian trajectory model when crossing a bridge is used for comprehensive identification of the social force behavior preferences of pedestrians crossing a bridge.

2. The comprehensive identification method for social force behavior preference of pedestrians crossing bridges according to claim 1 is characterized in that: The questionnaire survey in S1 includes two stages: preliminary survey and formal survey; The preliminary survey verified the reliability of the questionnaire through reliability analysis and validity analysis; The formal investigation was based on a multi-scenario setting, including static obstacles, dynamic obstacles, emergencies, and high-density crowd flow scenarios.

3. The comprehensive identification method for social force behavior preference of pedestrians crossing bridges according to claim 2 is characterized in that: The reliability analysis was tested by Cronbach's α coefficient; The validity analysis was tested by exploratory factor analysis.

4. The comprehensive identification method for social force behavior preference of pedestrians crossing bridges according to claim 1 is characterized in that: The working conditions in S2 include: a: Under the condition of an unobstructed bridge surface, select 5-15 people to cross the bridge surface, with a male-to-female ratio of 2:1 and an age range of 19-25 years old; b: When there are obstacles on the bridge deck, 5-15 people are selected, with a male-to-female ratio of 2:1 and an age range of 19-25 years old.

5. The comprehensive identification method of social force behavior preference of pedestrians crossing bridges according to claim 1 is characterized in that: In S3, the YOLOv11x target detection model is used to extract the position information of pedestrians from the video frames; the DeepSORT tracking algorithm is used to obtain the position changes of pedestrians between consecutive frames, thereby constructing the pedestrian motion trajectory.

Images