Intelligent car-oriented pedestrian risk state avoidance behavior three-dimensional kinematics quantitative analysis method and system

Abstract

The present application belongs to the field of intelligent automobile active safety technology, and discloses a three-dimensional kinematics quantitative analysis method and system for pedestrian dangerous state avoidance behavior of intelligent automobile. A virtual reality environment of human-vehicle conflict dangerous state scene is constructed, so that pedestrians perform street crossing task with street crossing gait, and three-dimensional motion data of all joints of pedestrians in the process of vehicle approaching is collected by using an inertial measurement unit. According to the spatial motion characteristics, the avoidance behavior is classified into four modes, i.e. forward avoidance, backward avoidance, avoidance in the direction of vehicle driving and in-situ reaction. The joint angle, angular velocity and angular acceleration of each joint at a preset analysis time are calculated. The three-dimensional kinematics parameters of different avoidance behavior modes are quantitatively compared and analyzed, and the angular velocity difference and angular acceleration difference are output. The present application can reveal the multi-joint coordinated motion mechanism of pedestrian dangerous state avoidance behavior, and provide a theoretical basis and data support for pedestrian behavior recognition and risk prediction of intelligent automobile advanced driving assistance system.

Description

Technical Field

[0001] This invention relates to the field of active safety technology for intelligent vehicles, specifically to a three-dimensional kinematic quantitative analysis method and system for pedestrian hazard avoidance behavior in intelligent vehicles, applicable to the identification and risk prediction of pedestrian hazard avoidance behavior in advanced driver assistance systems. Background Technology

[0002] In real-world traffic accident scenarios, pedestrians are more vulnerable to serious injury due to the lack of protective devices, making them typical vulnerable road users. According to the World Health Organization's 2023 "Global Status Report on Road Safety," pedestrians account for 21% of all road fatalities globally. A pedestrian's posture at the time of a collision significantly impacts their risk and severity of injury. Therefore, in-depth research into the behavioral characteristics of pedestrians in dangerous conflicts with vehicles is a crucial foundation for developing intelligent vehicle pedestrian protection systems and active and passive safety technologies.

[0003] Current active safety systems primarily rely on two-dimensional or overall image information of pedestrians in risk identification and collision avoidance strategy design, lacking in-depth analysis of the three-dimensional motion characteristics in hazardous situations. In reality, the relationship between pedestrians and vehicles is not a simple one-way avoidance relationship, but a dynamic interaction process influenced by multiple factors. When faced with potential conflict, pedestrians make avoidance decisions based on the vehicle's motion state, the surrounding environment, and their own judgment; therefore, their posture exhibits significant dynamic changes and uncertainties in three-dimensional space.

[0004] To understand the patterns of pedestrian avoidance behavior in pedestrian-vehicle conflicts, researchers have conducted extensive studies from multiple perspectives, including traffic accident analysis, virtual scenario testing, and pedestrian behavior recognition. Studies based on real traffic accidents or natural driving data have identified key factors influencing the severity of pedestrian accidents. Some studies have utilized virtual experimental testing methods to examine pedestrian avoidance behavior in dangerous situations under controlled environments. However, while these methods can describe the overall reactions and behaviors of pedestrians, they lack a systematic observation of the three-dimensional motion characteristics of various joints in the human body.

[0005] Regarding the application of active safety systems in intelligent vehicles, existing research has focused on constructing pedestrian behavior models and evaluation frameworks. However, existing models largely rely on the overall motion information of pedestrians, making it difficult to reflect the coordinated motion characteristics of all joints in a pedestrian's body under dangerous conditions. Overall, quantitative analysis of the three-dimensional kinematics of pedestrian avoidance behavior in pedestrian-vehicle conflict scenarios remains insufficient, and systematic research on the mechanisms of coordinated joint motion is still lacking. Summary of the Invention

[0006] This invention provides a three-dimensional kinematic quantitative analysis method and system for pedestrian hazard avoidance behavior in intelligent vehicles.

[0007] The technical problem to be solved by this invention is that, in the existing technology, there is a lack of systematic and quantitative analysis of the three-dimensional kinematics of pedestrian avoidance behavior in pedestrian-vehicle conflict scenarios, a lack of revelation of the kinematic differences in different avoidance behavior patterns, and a lack of systematic explanation of the multi-joint collaborative motion mechanism. This invention provides a method and system for quantitative analysis of the three-dimensional kinematics of pedestrian avoidance behavior in intelligent vehicles.

[0008] The technical solution of the present invention is as follows: A three-dimensional kinematic quantitative analysis method for pedestrian hazard avoidance behavior in intelligent vehicles includes the following steps: S1: Construct a virtual reality environment for a dangerous scenario involving a human-vehicle conflict. The virtual reality environment includes at least a vehicle model, a road environment model, and vehicles traveling according to a preset trajectory and speed.

[0009] Preferably, the pedestrian-vehicle conflict scenario is selected from at least one of the following four typical scenarios: the basic scenario of vehicles going straight and pedestrians crossing perpendicularly, the scenario of multiple vehicles driving in parallel and pedestrians crossing perpendicularly with obstructed vision, the scenario of multiple vehicles driving in parallel and pedestrians crossing in the middle with obstructed vision, and the scenario of vehicles turning right and pedestrians crossing perpendicularly with turning interaction.

[0010] S2: The pedestrian enters the virtual reality environment and performs the task of crossing the street with a gait. Inertial measurement units arranged in multiple parts of the pedestrian's body collect the three-dimensional motion data of the pedestrian's whole body joints in real time during the approach of the vehicle.

[0011] Preferably, the three-dimensional motion data of the whole body joints includes at least: motion data of the thorax in the flexion-extension direction, axial rotation direction, and lateral bending direction; motion data of the shoulder joint in the flexion-extension direction and abduction-adduction direction; motion data of the elbow joint in the flexion-extension direction and ulnar deviation-radial deviation direction; motion data of the hip joint in the flexion-extension direction and abduction-adduction direction; and motion data of the knee joint in the flexion-extension direction and abduction-adduction direction.

[0012] S3: Based on the spatial movement characteristics of pedestrians during the approach of vehicles, pedestrian avoidance behavior is classified into four typical modes: forward avoidance, backward avoidance, avoidance in the direction of vehicle travel, and insufficient reaction time.

[0013] Preferably, the classification is based on the following: when the pedestrian's center of mass displacement or lower limb stride is less than half the length of the pedestrian's stride, it is determined that the pedestrian's reaction is too slow; otherwise, based on the pedestrian's spatial displacement direction relative to its initial direction of travel, it is determined to avoid the situation by moving forward, backward, or in the direction of the vehicle's travel.

[0014] S4: For each avoidance behavior mode, calculate the three-dimensional kinematic parameters of each joint at the preset analysis time. The three-dimensional kinematic parameters include joint angle, joint angular velocity and joint angular acceleration.

[0015] Preferably, the preset analysis time is the common moment when all joint angles are within the range of their peak values.

[0016] S5: Perform quantitative comparative analysis on the three-dimensional kinematic parameters under different avoidance behavior modes, and output the joint three-dimensional kinematic difference characteristics between different avoidance behavior modes, wherein the difference characteristics include at least angular velocity difference and angular acceleration difference.

[0017] Preferably, the quantitative comparative analysis includes: comparing the distribution range and directional characteristics of each joint on the three-dimensional kinematic parameters under different avoidance behavior modes, so as to identify the kinematic differences between active avoidance behavior and inadequate on-the-spot reaction, as well as the kinematic differences between different active avoidance behavior modes.

[0018] Preferably, the method further includes step S6: generating pedestrian behavior recognition parameters based on the three-dimensional kinematic differences in joints among the various avoidance behavior patterns; the behavior recognition parameters include: angular velocity and angular acceleration of lower limb joints, used to determine the pedestrian's movement tendency during the avoidance process; angular velocity and angular acceleration characteristics of the thoracic cavity in the flexion or extension direction, used to assist in recognizing pedestrian avoidance behavior patterns; flexion and adduction angle characteristics of upper limb joints, used to generate high collision risk warning signals; the behavior recognition parameters are used to input into the advanced driver assistance system or active safety system of the intelligent vehicle.

[0019] A three-dimensional kinematic quantitative analysis system for pedestrian hazard avoidance behavior in intelligent vehicles, used to implement the method, includes: The virtual scene construction module is used to construct a virtual reality environment for pedestrian-vehicle conflict scenarios. The virtual reality environment includes at least a vehicle model, a road environment model, and vehicles traveling according to a preset trajectory and speed. The pedestrian-vehicle conflict scenarios are selected from at least one of the following four typical scenarios: a basic scenario of vehicles going straight and pedestrians crossing perpendicularly, a scenario of multiple vehicles driving side by side and pedestrians crossing perpendicularly with obstructed view, a scenario of multiple vehicles driving side by side and pedestrians crossing in the middle with obstructed view, and a scenario of vehicles turning right and pedestrians crossing perpendicularly with a turning interaction.

[0020] An inertial motion capture module includes inertial measurement units arranged at multiple parts of a pedestrian's body. These units are used to collect real-time three-dimensional motion data of the pedestrian's joints during vehicle approach when the pedestrian is performing a street-crossing gait in the virtual reality environment. The three-dimensional motion data includes at least the following: motion data of the thoracic cavity in the flexion-extension, axial rotation, and lateral bending directions; motion data of the shoulder joint in the flexion-extension and abduction-adduction directions; motion data of the elbow joint in the flexion-extension and ulnar-radial deviation directions; motion data of the hip joint in the flexion-extension and abduction-adduction directions; and motion data of the knee joint in the flexion-extension and abduction-adduction directions.

[0021] The behavior pattern classification module is used to classify pedestrian avoidance behavior into four typical patterns based on the spatial movement characteristics of pedestrians during the approach of vehicles: forward avoidance, backward avoidance, avoidance in the direction of vehicle travel, and insufficient reaction time.

[0022] The kinematic parameter calculation module is used to calculate the three-dimensional kinematic parameters of each joint at a preset analysis time for each avoidance behavior mode. The three-dimensional kinematic parameters include joint angle, joint angular velocity, and joint angular acceleration.

[0023] The difference feature output module is used to perform quantitative comparative analysis of the three-dimensional kinematic parameters under different avoidance behavior modes, and output the joint three-dimensional kinematic difference features between different avoidance behavior modes. The difference features include at least angular velocity difference and angular acceleration difference.

[0024] Preferably, the system further includes a parameter generation module for generating pedestrian behavior recognition parameters based on the three-dimensional kinematic differences in joints among the various avoidance behavior patterns. The behavior recognition parameters include: angular velocity and angular acceleration of the lower limb joints, used to determine the pedestrian's movement tendency during the avoidance process; angular velocity and angular acceleration characteristics of the thoracic cavity in the flexion or extension direction, used to assist in identifying pedestrian avoidance behavior patterns; and flexion and adduction angle characteristics of the upper limb joints, used to generate high collision risk warning signals. The behavior recognition parameters are input to the advanced driver assistance system or active safety system of the intelligent vehicle.

[0025] The beneficial effects of this invention are as follows: First, it enables a systematic and quantitative analysis of pedestrian avoidance behavior. By integrating high-fidelity virtual scenes with an inertial motion capture system, a virtual-real coupled experimental platform is constructed. Under safe and controllable conditions, the interaction process between pedestrians and vehicles is realistically reproduced. The system collects three-dimensional motion data of the entire body joints of pedestrians in dangerous scenarios, enabling a systematic analysis of pedestrian avoidance behavior from the overall to the joint level, and from qualitative to quantitative analysis.

[0026] Second, a classification system for four typical pedestrian avoidance behavior patterns was established. Pedestrian avoidance behavior was summarized into four typical patterns: forward avoidance, backward avoidance, avoidance in the direction of vehicle travel, and insufficient on-the-spot reaction. Quantitative classification criteria for each pattern were provided. This classification system provides a basic framework for the standardized description of pedestrian avoidance behavior and for subsequent research.

[0027] Third, the study revealed the kinematic differences among different avoidance behavior patterns. A quantitative comparison of the distribution characteristics and differences of joint angles, angular velocities, and angular accelerations in different avoidance behavior patterns revealed that the differences in pedestrian joint angles were relatively small, while the differences in angular velocities and angular accelerations were more significant. Active avoidance behavior, on the whole, exhibited greater angular velocities and angular accelerations, demonstrating faster response speeds and stronger explosive movements.

[0028] Fourth, it revealed the multi-joint coordinated movement mechanism. From the perspective of three-dimensional joint kinematics, it revealed the overall movement mechanism of pedestrian avoidance behavior in dangerous situations, and found that pedestrian avoidance behavior exhibits a multi-joint coordinated movement mechanism in which the lower limbs dominate the displacement and the thorax and upper limbs work together to maintain stability.

[0029] Fifth, it provides operable technical solutions for active safety systems in intelligent vehicles. It explicitly proposes operable technical indicators for transforming the three-dimensional kinematic features of joints into pedestrian behavior recognition and risk prediction in advanced driver assistance systems for intelligent vehicles. Specific application schemes include prioritizing the extraction of dynamic change trends of lower limb joints to determine avoidance tendencies, utilizing the angular velocity and angular acceleration characteristics of the thoracic cavity to assist in identifying avoidance behavior patterns, and using upper limb defensive postures as high-collision-risk signals. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein: Figure 1 This is a schematic diagram of four typical pedestrian crossing hazards in this invention.

[0031] Figure 2 This is a schematic diagram showing the frequency distribution of different avoidance behavior patterns in this invention.

[0032] Figure 3 This is a schematic diagram illustrating the range of motion of the thoracic joint angle under different avoidance behavior modes in this invention.

[0033] Figure 4 This is a schematic diagram illustrating the range of motion of the shoulder joint angle under different avoidance behavior modes in this invention.

[0034] Figure 5 This is a schematic diagram illustrating the range of motion of the elbow joint angle under different avoidance behavior modes in this invention.

[0035] Figure 6 This is a schematic diagram illustrating the range of motion of the hip joint angle under different avoidance behavior modes in this invention.

[0036] Figure 7 This is a schematic diagram illustrating the range of motion of the knee joint angle under different avoidance behavior modes in this invention. Detailed Implementation

[0037] The specific embodiments of the present invention will be described in detail below with reference to the technical solution of the present invention. It should be understood that the following embodiments are only used to illustrate the present invention and are not intended to limit the scope of protection of the present invention. Without departing from the concept of the present invention, those skilled in the art can make equivalent substitutions or appropriate adjustments to the relevant structures, connection methods, parameter ranges and data processing methods.

[0038] This invention provides a three-dimensional kinematic quantitative analysis method for pedestrian hazard avoidance behavior in intelligent vehicles, comprising the following steps: S1: Construct a virtual reality environment for a dangerous scenario of human-vehicle conflict, wherein the virtual reality environment includes at least a vehicle model, a road environment model, and vehicles traveling according to a preset trajectory and speed; S2: The pedestrian enters the virtual reality environment and performs the crossing task with a crossing gait. Inertial measurement units arranged in multiple parts of the pedestrian's body collect the three-dimensional motion data of the pedestrian's whole body joints in real time during the vehicle approach process. S3: Based on the spatial movement characteristics of pedestrians during the approach of vehicles, pedestrian avoidance behavior is classified into four typical modes: forward avoidance, backward avoidance, avoidance in the direction of vehicle travel, and insufficient reaction time. S4: For each avoidance behavior mode, calculate the three-dimensional kinematic parameters of each joint at the preset analysis time. The three-dimensional kinematic parameters include joint angle, joint angular velocity and joint angular acceleration. S5: Perform quantitative comparative analysis on the three-dimensional kinematic parameters under different avoidance behavior modes, and output the joint three-dimensional kinematic difference characteristics between different avoidance behavior modes, wherein the difference characteristics include at least angular velocity difference and angular acceleration difference.

[0039] In step S1, the pedestrian-vehicle conflict scenario is selected from at least one of the following four typical scenarios: the basic scenario of vehicles going straight and pedestrians crossing perpendicularly, the scenario of multiple vehicles driving in parallel and pedestrians crossing perpendicularly with obstructed vision, the scenario of multiple vehicles driving in parallel and pedestrians crossing in the middle with obstructed vision, and the scenario of vehicles turning right and pedestrians crossing perpendicularly with turning interaction.

[0040] In step S2, the three-dimensional motion data of the whole body joints includes at least: Motion data of the thorax in the flexion-extension direction, axial rotation direction, and lateral bending direction; Movement data of the shoulder joint in the flexion-extension and abduction-adduction directions; Motion data of the elbow joint in the flexion-extension direction and the ulnar and radial deviation directions; Hip joint motion data in flexion-extension and abduction-adduction directions; And the motion data of the knee joint in the flexion-extension direction and the abduction-adduction direction.

[0041] In step S3, the classification is based on the following: when the pedestrian's center of mass displacement or lower limb stride is less than half the length of the pedestrian's stride, it is determined that the pedestrian's reaction is too slow; otherwise, based on the pedestrian's spatial displacement direction relative to its initial direction of travel, it is determined to avoid the vehicle by moving forward, backward, or in the direction of travel.

[0042] The method further includes step S6: generating pedestrian behavior recognition parameters based on the three-dimensional kinematic differences in joints among the various avoidance behavior patterns; the behavior recognition parameters include: angular velocity and angular acceleration of lower limb joints, used to determine the pedestrian's movement tendency during the avoidance process; angular velocity and angular acceleration characteristics of the thoracic cavity in the flexion or extension direction, used to assist in recognizing pedestrian avoidance behavior patterns; flexion angle and adduction angle characteristics of upper limb joints, used to generate high collision risk warning signals; the behavior recognition parameters are used to input into the advanced driver assistance system or active safety system of intelligent vehicles.

[0043] This invention also provides a three-dimensional kinematic quantitative analysis system for pedestrian hazard avoidance behavior in intelligent vehicles, used to implement the method, including: The virtual scene construction module is used to construct a virtual reality environment for a pedestrian-vehicle conflict scenario. The virtual reality environment includes at least a vehicle model, a road environment model, and vehicles traveling according to a preset trajectory and speed. The pedestrian-vehicle conflict scenario is selected from at least one of the following four typical scenarios: a basic scenario of vehicles going straight and pedestrians crossing perpendicularly, a scenario of multiple vehicles driving in parallel and pedestrians crossing perpendicularly with obstructed view, a scenario of multiple vehicles driving in parallel and pedestrians crossing in the middle with obstructed view, and a scenario of vehicles turning right and pedestrians crossing perpendicularly. An inertial motion capture module includes inertial measurement units arranged at multiple parts of a pedestrian's body. These units are used to collect real-time three-dimensional motion data of the pedestrian's entire body joints during vehicle approach when the pedestrian is performing a street-crossing task in the virtual reality environment using a street-crossing gait. The three-dimensional motion data of the entire body joints includes at least the following: Motion data of the thorax in the flexion-extension direction, axial rotation direction, and lateral bending direction; Movement data of the shoulder joint in the flexion-extension and abduction-adduction directions; Motion data of the elbow joint in the flexion-extension direction and the ulnar and radial deviation directions; Hip joint motion data in flexion-extension and abduction-adduction directions; And motion data of the knee joint in the flexion-extension direction and the abduction-adduction direction; The behavior pattern classification module is used to classify pedestrian avoidance behavior into four typical patterns based on the spatial movement characteristics of pedestrians during the approach of vehicles: forward avoidance, backward avoidance, avoidance in the direction of vehicle travel, and insufficient reaction time. The kinematic parameter calculation module is used to calculate the three-dimensional kinematic parameters of each joint at a preset analysis time for each avoidance behavior mode. The three-dimensional kinematic parameters include joint angle, joint angular velocity, and joint angular acceleration. The difference feature output module is used to perform quantitative comparative analysis of the three-dimensional kinematic parameters under different avoidance behavior modes, and output the joint three-dimensional kinematic difference features between different avoidance behavior modes. The difference features include at least angular velocity difference and angular acceleration difference.

[0044] The classification criteria of the behavior pattern classification module are as follows: when the pedestrian's center of mass displacement or lower limb stride is less than half the length of the pedestrian's stride, it is determined that the pedestrian's reaction is too slow; otherwise, based on the spatial displacement direction of the pedestrian relative to its initial direction of travel, it is determined to avoid the vehicle by moving forward, avoiding the vehicle by moving backward, or avoiding the vehicle's direction of travel.

[0045] The preset analysis time in the kinematic parameter calculation module is the common moment when the angles of each joint are all within the range of their peak values.

[0046] The quantitative comparative analysis in the difference feature output module includes: comparing the distribution range and directional characteristics of each joint on the three-dimensional kinematic parameters under different avoidance behavior modes, so as to identify the kinematic differences between active avoidance behavior and inadequate on-the-spot reaction, as well as the kinematic differences between different active avoidance behavior modes.

[0047] The system further includes a parameter generation module for generating pedestrian behavior recognition parameters based on the three-dimensional kinematic differences in joints among the various avoidance behavior patterns. These behavior recognition parameters include: angular velocity and angular acceleration of the lower limb joints, used to determine the pedestrian's movement tendency during avoidance; angular velocity and angular acceleration characteristics of the thoracic cavity in the flexion or extension directions, used to assist in identifying pedestrian avoidance behavior patterns; and flexion and adduction angle characteristics of the upper limb joints, used to generate high collision risk warning signals. These behavior recognition parameters are input into the advanced driver assistance system or active safety system of the intelligent vehicle.

[0048] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. The following embodiments are organized according to a logical progression: experimental platform construction, experimental data acquisition, data processing and parameter calculation, behavioral pattern classification, kinematic parameter comparison and analysis, revelation of multi-joint coordination mechanisms, and active safety applications. The embodiments are interconnected and progressively build upon each other, collectively constituting the complete technical solution of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0049] I. Construction of a virtual-real coupling experimental platform; This embodiment details the construction method of the virtual-real coupling experimental platform used in this invention.

[0050] (a) Virtual scene construction; This embodiment utilizes the CARLA autonomous driving simulation platform to construct a virtual traffic scenario. CARLA is an open virtual testing platform widely used in traffic scenario research. Based on typical pedestrian-vehicle conflict scenarios, the positions of pedestrians and vehicles, as well as road environment parameters, are set. To ensure the realism of the pedestrian-vehicle interaction process, combined with urban road characteristics, vehicle speeds are uniformly set through virtual platform preview: 40 km / h for straight sections and 30 km / h for turning sections. The four typical pedestrian crossing hazard virtual scenarios after completion are shown below. Figure 1 As shown.

[0051] The four typical scenarios are as follows: Scenario 1: Basic scenario of vehicles going straight and pedestrians crossing perpendicularly. This scenario simulates a situation where pedestrians cross a lane perpendicularly without obstruction, resulting in a conflict with vehicles going straight.

[0052] Scenario 2: Multi-vehicle parallel crossing - pedestrian perpendicular crossing vision obstruction scenario. This scenario simulates a high-risk situation where pedestrians are crossing perpendicularly in a multi-lane road, and vehicles in adjacent lanes obstruct their view.

[0053] Scenario 3: Multiple vehicles driving side by side - pedestrians crossing the middle of a workshop with obstructed view. This scenario simulates a complex conflict situation where pedestrians, while waiting between two vehicles, must cross the workshop and simultaneously face obstructed view.

[0054] Scenario 4: Turning interaction scenario between a vehicle turning right and a pedestrian crossing perpendicularly. This scenario simulates a high-risk situation where a pedestrian crossing perpendicularly at an intersection interacts with a vehicle turning right.

[0055] The four scenarios mentioned above cover the most typical and dangerous human-vehicle interaction situations on urban roads. By selecting at least one scenario for testing, data on pedestrian avoidance behavior under different conflict conditions can be effectively obtained.

[0056] (ii) Configuration of the inertial motion capture system; In this embodiment, Awinda inertial measurement units (IMUs) are deployed at 17 locations on the subject's body. These IMUs are used to collect three-dimensional motion data of the subject's joints during a street-crossing task in a virtual environment. Real-time synchronization and posture data acquisition are achieved using Xsens software, enabling dynamic tracking of each body part. This motion capture system is unaffected by visual occlusion and can stably and accurately record the motion process.

[0057] (III) The workflow of the virtual-real coupling experimental platform; The workflow of the virtual-real coupling experimental platform is as follows: First, a virtual scene is built based on the CARLA simulation platform. The virtual scene includes vehicle models, road environment models, and vehicles traveling according to preset trajectories and speeds.

[0058] Secondly, the subjects wore VR headsets, which created an immersive street-crossing experience for them. Simultaneously, inertial measurement units were deployed at 17 locations on the subjects' bodies.

[0059] Then, the participants completed a street crossing task in a virtual environment. No specific guidance was provided during the experiment; participants wore VR devices and independently perceived the surrounding traffic environment and oncoming vehicle information, ensuring that pedestrian reactions were autonomous and unpredictable.

[0060] Finally, the vehicles in the scenario travel at a pre-set speed and trajectory, creating a conflict with the subjects, and the subjects' motion trajectory and joint movement data are collected synchronously throughout the process.

[0061] II. Experimental Design and Data Acquisition; This embodiment details the specific experimental scheme for data acquisition using the above-described experimental platform.

[0062] (a) Participants in the experiment; This experiment recruited 18 volunteers aged 20 to 28. This age group represents adults with relatively stable physical functions and reaction abilities, and is also a typical group with high participation in urban traffic. Both male and female participants were included to eliminate the bias caused by gender differences. There were also some differences in height and weight among the participants. Before the experiment, all participants underwent standardized testing to familiarize themselves with the VR equipment and adapt to the virtual traffic scenario, minimizing the impact of unfamiliarity and preconceived notions on pedestrian decision-making.

[0063] (II) Experimental Design; In the experiment, participants wore VR devices and an inertial motion capture system. After system debugging and human model calibration, they completed a street crossing task in a virtual environment. No specific avoidance methods were guided during the experiment; participants, wearing VR devices, perceived the surrounding traffic environment and oncoming vehicle information independently, ensuring the pedestrian's reaction was autonomous and unpredictable. Vehicles in the scenario traveled at pre-set speeds and trajectories, colliding with the participants. The participants' motion trajectories and joint movement data were collected synchronously throughout the experiment. Each participant performed two replicate experiments in each of the four virtual scenarios, obtaining a total of 144 sets of experimental data (18 participants multiplied by 4 scenarios multiplied by 2).

[0064] III. Definition and calculation of three-dimensional kinematic parameters of joints; This embodiment details the definition method and calculation process of the three-dimensional kinematic parameters of the joint used in this invention.

[0065] (a) Analysis of joint selection; Based on the structural characteristics of human movement, this embodiment selects nine joints, including the chest, shoulder, elbow, hip, and knee, as the analysis objects.

[0066] (ii) Definition of joint degrees of freedom; After selecting the pedestrian's walking direction as the positive Y-axis and the vehicle's driving direction as the positive X-axis, the rotation direction of each joint angle is defined as follows: The degrees of freedom of movement of the thoracic cavity include: flexion and extension, axial rotation, and lateral bending.

[0067] The range of motion of the shoulder joint includes flexion and extension, abduction and adduction.

[0068] The range of motion of the elbow joint includes flexion and extension, ulnar deviation, and radial deviation.

[0069] The degrees of freedom of movement of the hip joint include: flexion and extension, abduction and adduction.

[0070] The knee joint's degrees of freedom of movement include: flexion and extension, abduction and adduction.

[0071] (iii) Determining the preset analysis time; By observing the subject's avoidance actions frame by frame, the common moment when all joint angles are within the range of their peak values ​​was determined. The joint angles at this moment were extracted, and angular velocity and angular acceleration were calculated. The range of the peak values ​​refers to the interval where each joint angle value is within ±10% of its peak value. This moment represents the extreme node of posture adjustment during pedestrian avoidance, significantly highlighting the differences in pedestrian actions between different avoidance modes during the decision-making and execution of avoidance.

[0072] (iv) Calculation methods for three-dimensional kinematic parameters; Joint angle calculation: Based on the quaternion data of each joint collected by the inertial measurement unit, the angle values ​​of each joint in the corresponding degree of freedom are calculated through coordinate system transformation. Joint angles can effectively characterize the body posture of a pedestrian in dangerous scenarios.

[0073] Calculation of joint angular velocity: The measured value of angular velocity can be directly output by the inertial measurement unit, reflecting the speed of limb movement and the pedestrian's response speed to sudden risks.

[0074] Calculation of joint angular acceleration: Angular acceleration is obtained by differentiating the angular velocity over time.

[0075] The direction of angular velocity and angular acceleration can reveal the next movement tendency of a joint, while the coordinated movement of multiple joints can characterize the next avoidance action of the posture.

[0076] IV. Classification of pedestrian avoidance behavior patterns; This embodiment details the method for classifying pedestrian avoidance behavior patterns in this invention.

[0077] (a) Classification of behavioral patterns; Analyzing the animations captured by the motion capture system, and based on the pedestrian's spatial motion relative to the initial direction of travel as the vehicle approaches, the pedestrian's avoidance behavior is divided into four basic types: Avoiding a vehicle: When a pedestrian senses a vehicle approaching, they accelerate forward in their initial direction of travel to cross the vehicle's path.

[0078] Avoiding a vehicle by moving backward: When a pedestrian senses a vehicle approaching, they overcome the inertia of their initial direction of motion and retreat backward to return to a safe area.

[0079] Avoiding a vehicle by moving in the direction of travel: When a pedestrian senses a vehicle approaching, they move in the direction the vehicle is traveling to avoid it.

[0080] Insufficient on-the-spot reaction: Pedestrians fail to make effective spatial displacement to avoid the approaching vehicle after sensing its approach.

[0081] (ii) Quantitative criteria for classification; The displacement of the pedestrian's center of gravity and the direction of their lower limb stride after the vehicle enters the subject's risk perception range are used as the main criteria for classification. Stride length refers to the straight-line distance between two consecutive heel strike points on the same side during normal street crossing; the average of multiple measurements is taken as the pedestrian's stride length. When the pedestrian's center of gravity displacement or lower limb stride is less than half of their stride length, the behavior is considered an ineffective avoidance tactic and is classified as insufficient on-the-spot reaction.

[0082] Otherwise, based on the pedestrian's spatial displacement direction relative to their initial direction of travel, it is determined whether to avoid the vehicle by moving forward, backward, or in the direction of travel.

[0083] Specifically: if the pedestrian's center of gravity displacement direction is the same as the initial direction of travel, and the displacement amplitude is greater than half the stride length, it is determined to avoid the vehicle forward; if the pedestrian's center of gravity displacement direction is opposite to the initial direction of travel, and the displacement amplitude is greater than half the stride length, it is determined to avoid the vehicle backward; if the pedestrian's center of gravity displacement direction is the same as the vehicle's direction of travel, and the displacement amplitude is greater than half the stride length, it is determined to avoid the vehicle's direction of travel.

[0084] (iii) Distribution characteristics of behavioral patterns; Statistical analysis of 144 experimental samples revealed significant diversity in pedestrian avoidance behaviors in dangerous situations. The frequency distribution of each avoidance behavior showed that forward avoidance occurred 42 times, avoidance towards the direction of vehicle travel occurred 48 times, backward avoidance occurred 32 times, and insufficient reaction time occurred 22 times. The frequency distribution of different avoidance behavior patterns is as follows: Figure 2 As shown.

[0085] V. Comparison of kinematic parameters for different avoidance behavior patterns; This embodiment details the comparative analysis method and results of the three-dimensional kinematic parameters of each joint under different avoidance behavior modes.

[0086] (a) Comparative analysis of joint angles; Under different avoidance behavior patterns, the angular distribution of the thoracic cavity in the flexion-extension, axial rotation, and lateral bending directions showed little difference, and the overall distribution characteristics were quite similar.

[0087] Regarding chest flexion and extension, the chest angles for all avoidance behaviors were negative, indicating a bent-over posture. Among them, the flexion angles were largest for forward avoidance and avoidance in the direction of vehicle travel, followed by backward avoidance, while the flexion angle was smallest for inability to react in place.

[0088] Regarding axial rotation, the rotation angles of various avoidance behaviors are mostly positive, manifesting as a leftward rotation posture, i.e., facing the direction of the vehicle, while the leftward rotation amplitude for backward avoidance is the smallest.

[0089] Regarding lateral bending, the median bending angle for each avoidance behavior is close to zero, indicating that there is no obvious tendency for lateral bending overall. However, the data distribution range is large, with the largest lateral tilting activity observed when avoiding forward movement and avoiding movement in the direction of vehicle travel.

[0090] Range of motion of the thoracic joint angle under different avoidance behavior patterns, such as Figure 3 As shown.

[0091] Under different avoidance behavior patterns, the range of motion of the shoulder joint is relatively small and the overall distribution characteristics are quite similar, and the elbow joint also shows a similar pattern.

[0092] Regarding shoulder flexion and extension, the angles of most avoidance behaviors were positive, manifesting as flexion movements, with the smallest flexion range observed in backward avoidance. Regarding shoulder abduction and adduction, the angles of the left shoulder joint were almost entirely positive, while the angles of the right shoulder joint were almost entirely negative, indicating that the shoulder joint response was generally abduction, with the range mostly within 40 degrees.

[0093] Range of motion of shoulder joint angle under different avoidance behavior patterns, such as Figure 4 As shown.

[0094] Regarding elbow flexion and extension, the elbow angles for all avoidance behaviors were almost all positive, indicating flexion movements with a large amplitude; the median was over 50 degrees, and the maximum was no more than 140 degrees. Regarding ulnar and radial deviation of the elbow joint, the left elbow angles for all avoidance behaviors were more negative, while the right elbow angles were more positive; that is, the elbow joint response was generally ulnar, with an amplitude within ±50 degrees.

[0095] Range of motion of elbow joint angle under different avoidance behavior patterns, such as Figure 5 As shown.

[0096] The angle range of the hip joint did not differ significantly under different avoidance behavior patterns, and the distribution characteristics were basically the same. The knee joint also showed similar characteristics.

[0097] Regarding hip flexion and extension, almost all avoidance behaviors showed positive angles, exhibiting flexion movements with amplitudes within 80 degrees. Forward avoidance and avoidance in the direction of vehicle travel showed the largest amplitudes, followed by backward avoidance, while on-the-spot reactions showed the smallest amplitudes. Regarding hip abduction and adduction, the left hip joint angles were more frequently positive, while the right hip joint angles were more frequently negative, indicating that the hip joint response was generally abduction, with amplitudes within ±40 degrees.

[0098] Range of motion of hip joint angle under different avoidance behavior patterns, such as Figure 6 As shown.

[0099] Regarding knee flexion and extension, the knee angles for all avoidance behaviors were almost all negative, manifesting as shin flexion with an amplitude of less than 90 degrees. Among them, the knee flexion amplitude was the largest for backward avoidance, while the amplitude was the smallest for on-the-spot reactions. Regarding knee abduction and adduction, the knee angles for all avoidance behaviors were very small and mostly close to zero degrees, so the range of motion of the knee joint in abduction and adduction was extremely small.

[0100] Range of motion of knee joint angle under different avoidance behavior patterns, such as Figure 7 As shown.

[0101] (II) Comparative analysis of joint angular velocities; Statistical analysis of the three-dimensional angular velocities of each joint shows that the magnitude of the angular velocity reflects the speed of the limb's response, while the direction of the angular velocity characterizes the tendency of the joint to move in its degrees of freedom.

[0102] Overall, among the various avoidance behavior patterns, the upper limb joints, namely the shoulder and elbow joints, exhibited the highest angular velocities, demonstrating the fastest response. The lower limb joints, namely the hip and knee joints, followed. The thoracic cavity exhibited the lowest angular velocity, resulting in the slowest response.

[0103] Regarding thoracic flexion and extension, the angular velocity for backward avoidance is positive and relatively large, with a faster response in the backward extension direction. Other avoidance behaviors have relatively smaller angular velocities and a slower response in the flexion direction.

[0104] Regarding axial rotation of the thoracic cavity, the angular velocity of each avoidance behavior is relatively small overall, and its joint response is slow.

[0105] Regarding lateral flexion of the ribcage, the angular velocities of all avoidance maneuvers are positive, with a tendency to flex to the right. The angular velocity is relatively large for forward avoidance and relatively small for backward avoidance.

[0106] Regarding shoulder flexion and extension, the angular velocities of each avoidance behavior are relatively large, and the response speed is relatively fast. Among them, the angular velocity of backward avoidance is negative, which is manifested as a tendency for the shoulder to swing backward and extend.

[0107] In terms of shoulder abduction and adduction, the angular velocities of each avoidance behavior are relatively small, reflecting a relatively limited response speed in this degree of freedom.

[0108] In terms of elbow flexion and extension, the angular velocities of all avoidance behaviors are relatively large, and are mainly positive, with a faster response in the flexion direction.

[0109] Regarding ulnar and radial deviation of the elbow joint, the angular velocities of each avoidance behavior are relatively small, and the response speed is relatively limited.

[0110] Regarding hip flexion and extension, the angular velocities of various avoidance behaviors are mostly negative, with a tendency for the legs to extend backward. Among them, the angular velocity of backward avoidance is the largest, and the response in the backward extension direction is the fastest.

[0111] In terms of hip abduction and adduction, the angular velocities of the various avoidance behaviors are relatively small, with the left hip showing positive values ​​and the right hip mostly showing negative values, indicating an abduction trend.

[0112] Regarding knee flexion and extension, the angular velocity of backward avoidance is the largest and positive, with the fastest response in the extension direction. The angular velocity of forward avoidance is relatively large and negative, with a faster response in the flexion direction. Avoidance in the direction of vehicle movement and reaction time in place are both slower, resulting in a smaller angular velocity and limited response speed.

[0113] In terms of knee abduction and adduction, the angular velocities and differences between various avoidance behaviors are relatively small, and the response speed is limited.

[0114] (III) Comparative analysis of joint angular acceleration; The angular acceleration characteristics of each joint also differ under different avoidance behavior patterns. The magnitude of angular acceleration can reflect the explosiveness of limb movement, while the direction of angular acceleration characterizes the trend of joint movement changes in its degrees of freedom.

[0115] Overall, the angular acceleration of the lower limb joints is greater than that of the thorax and upper limbs, indicating that the lower limbs exhibit stronger explosiveness in avoidance.

[0116] Regarding thoracic flexion and extension, the angular acceleration for forward avoidance and avoidance in the direction of vehicle travel is negative, accelerating in the flexion direction. The angular acceleration for backward avoidance and failure to react in place is positive, accelerating in the extension direction. Among these, the angular acceleration for forward avoidance is the largest, reflecting a high-intensity, explosive movement response.

[0117] Regarding axial rotation of the thoracic cavity, the angular acceleration when avoiding the vehicle's direction of travel is negative and relatively large, while the acceleration in the rightward rotation direction is more explosive.

[0118] In terms of lateral flexion of the ribcage, the angular acceleration for dodging forward and backward is negative and relatively large, while the acceleration in the leftward flexion direction is more explosive.

[0119] Regarding shoulder flexion and extension, the angular acceleration is negative when avoiding the vehicle's direction or when unable to react in place, while it accelerates when extending backward. Among these, the angular acceleration is the greatest when avoiding the vehicle's direction, resulting in the most explosive movement.

[0120] In terms of shoulder abduction and adduction, the angular acceleration and differences in each avoidance behavior are relatively small, and the explosiveness is relatively limited.

[0121] Regarding elbow flexion and extension, the angular acceleration of each avoidance behavior is mostly negative, accelerating in the direction of extension. Among them, the angular acceleration is larger and more explosive when dodging backward or reacting in place.

[0122] Regarding ulnar and radial deviation of the elbow joint, the left elbow angular acceleration is negative and the right elbow is positive for all avoidance behaviors, with acceleration in the adduction direction. Among them, the angular acceleration for slow reaction in place and avoidance in the direction of vehicle travel is relatively large and explosive.

[0123] In terms of hip flexion and extension, the angular acceleration of each avoidance behavior is mostly positive, with the leg accelerating in the flexion direction.

[0124] Regarding hip abduction and adduction, the left hip angle acceleration is mostly negative and the right hip is positive for all avoidance behaviors, with acceleration in the abduction direction. Among them, the acceleration is the greatest and most explosive when dodging forward.

[0125] Regarding knee flexion and extension, the angular accelerations for all avoidance behaviors are relatively large, especially for forward avoidance and avoidance in the direction of vehicle travel, where the angular accelerations are negative, with a strong explosive acceleration in the flexion direction. The angular accelerations for backward avoidance and failure to react in place are relatively small and mostly positive, with acceleration in the forward extension direction.

[0126] Regarding knee abduction and adduction, the angular accelerations for all avoidance behaviors were relatively large, especially for forward avoidance and avoidance towards the direction of vehicle travel. Among these behaviors, the knee joint angular accelerations were mostly positive, with the left knee accelerating outward and the right knee accelerating inward, and the gait exhibiting characteristics of avoiding the direction of vehicle travel.

[0127] VI. Analysis of the mechanism of multi-joint coordinated movement; Based on the comparative analysis results of the above three-dimensional kinematic parameters, this embodiment details the multi-joint coordinated motion mechanism of pedestrian danger avoidance behavior disclosed in this invention.

[0128] (a) The dominant driving role of the lower limbs; The lower limb joints, especially the knee joint, showed significant differences in angular velocity and angular acceleration under different avoidance behaviors.

[0129] Specifically, when dodging forward or in the direction of vehicle travel, the knee joint exhibits a large flexion angular velocity and angular acceleration in order to quickly push off the ground and take a forward or lateral step.

[0130] When dodging backward, the knee joint exhibits a large extension angular velocity and a significant angular acceleration. Simultaneously, the hip joint exhibits a large extension angular velocity to facilitate a rapid backward step.

[0131] This directional and explosive response characteristic confirms the dominant driving role of the lower limbs in avoidance behavior, that is, the lower limbs are responsible for initiating the reaction and spatial displacement, and can respond rapidly in the time dimension.

[0132] (ii) The postural stabilizing function of the thorax; The rib cage primarily maintains postural stability. The joint angular velocities and angular accelerations of the rib cage are generally low, which is related to its large mass and the need to maintain a stable center of gravity. As the body's center of gravity, the rib cage must prioritize overall balance and stability, thus limiting its motion response.

[0133] However, the thoracic rib cage exhibits significant directional differences in flexion and extension movements, which, in conjunction with the lower limbs, help maintain body stability. The angular velocity and angular acceleration of the thoracic rib cage during avoidance behavior show marked directional differences, especially in flexion and extension, which can be used to assist in identifying pedestrian avoidance behavior patterns.

[0134] (iii) Defensive postures of the upper limbs; The upper limbs exhibit a distinctly defensive posture. In all avoidance maneuvers, the shoulder and elbow joints generally show a tendency to flex and adduct, with relatively high angular velocities.

[0135] Although the upper limbs do not drive the overall displacement of a pedestrian, they can protect vital areas such as the head and chest through flexion and adduction. In addition, the relatively large range of motion of the upper limb joints when there is insufficient time to react in place also confirms that when pedestrians cannot avoid danger, they will take self-protective measures to reduce the risk of injury.

[0136] (iv) Summary of multi-joint coordinated movement mechanisms; In summary, the multi-joint movement mechanism of pedestrian avoidance behavior can be summarized as a pattern in which the lower limbs drive displacement, while the thorax and upper limbs maintain postural stability.

[0137] VII. Applications in intelligent vehicle active safety systems; This embodiment details the application method of the present invention in intelligent vehicle advanced driver assistance systems and pedestrian protection systems.

[0138] (a) Generation of pedestrian behavior recognition parameters; Based on the differences in joint three-dimensional kinematics among the various avoidance behavior patterns, pedestrian behavior recognition parameters are generated.

[0139] The recognition parameter generation module is implemented as follows: It takes the three-dimensional kinematic differences in joints among the various avoidance behavior patterns as input, and maps the kinematic parameters of each joint to corresponding behavior recognition parameters using preset classification threshold judgment rules. Specifically, when the angular velocity of the lower limb joint is greater than a first preset threshold and the angular acceleration is greater than a second preset threshold, it is determined to be an active avoidance behavior, and recognition features are generated to determine the pedestrian's movement tendency during the avoidance process. When the direction of the angular velocity or angular acceleration of the thorax shows a significant change in flexion or extension, recognition features are generated to assist in identifying pedestrian avoidance behavior patterns. When the flexion angle of the upper limb joint is greater than a third preset threshold and the adduction angle is greater than a fourth preset threshold, recognition features are generated to generate a high collision risk warning signal.

[0140] The behavior recognition parameters include: First, the angular velocity and angular acceleration of the lower limb joints are used to determine the pedestrian's movement tendency during the avoidance process.

[0141] Second, the angular velocity and angular acceleration characteristics of the thoracic cavity in the flexion or extension direction are used to help identify pedestrian avoidance behavior patterns.

[0142] Third, the flexion and adduction angles of the upper limb joints are used to generate high collision risk warning signals.

[0143] (ii) Application strategies in advanced driver assistance systems; The behavior recognition parameters are input into the advanced driver assistance system or active safety system of the intelligent vehicle.

[0144] Specific application strategies include: First, when the vehicle recognizes that the pedestrian's lower limb joints are moving forward or in the direction of the vehicle's travel, it determines that the pedestrian is actively avoiding the collision. The system can then plan a reasonable collision avoidance strategy in advance based on the pedestrian's avoidance direction and speed.

[0145] Second, when a vehicle identifies a pedestrian's chest showing a significant difference in flexion or extension direction, it can be used to help determine the pedestrian's avoidance behavior pattern and improve the accuracy of risk prediction.

[0146] Third, when the vehicle recognizes that a pedestrian's upper limbs are in a defensive posture of flexion and adduction, it can send a risk warning to the pedestrian protection system and promptly trigger a pedestrian collision warning or automatic emergency braking.

[0147] Fourth, when a vehicle detects a failure to react in place, i.e., the pedestrian has not taken effective spatial avoidance measures, a high-priority instruction should be sent to trigger braking in advance or to adopt a more aggressive avoidance strategy to minimize the risk of collision and injury.

[0148] (III) Application examples in different scenarios; Example 1: Application in a basic scenario. In the basic scenario of a vehicle traveling straight and a pedestrian crossing perpendicularly, the pedestrian enters the crosswalk from the roadside, and the vehicle travels in a straight line. When the vehicle approaches, the inertial measurement unit collects real-time three-dimensional motion data of the pedestrian's entire body joints. The behavior pattern classification module determines the avoidance behavior pattern based on the direction of the pedestrian's center of mass displacement. If it is determined to be a forward avoidance, the kinematic parameter calculation module calculates the angular velocity and angular acceleration of the lower limb joints, and the difference feature output module outputs the difference in angular velocity and angular acceleration. The recognition parameter generation module generates behavior recognition parameters and inputs them into the advanced driver assistance system, which then triggers the corresponding collision avoidance strategy.

[0149] Example 2: Application in a Scene of Visual Obstruction. In a scene of multiple vehicles driving side-by-side and pedestrians crossing perpendicularly, vehicles in adjacent lanes obstruct the pedestrian's view. Since the inertial motion capture system is unaffected by visual obstruction, it can still stably collect three-dimensional motion data of the pedestrian's entire body joints. By comparing the differences in joint angular velocity and angular acceleration under different avoidance behavior modes, the parameter generation module generates behavior recognition parameters, providing an accurate basis for advanced driver assistance systems to judge pedestrian avoidance behavior.

[0150] Example 3: Application in a Turning Interaction Scenario. In a turning interaction scenario where a vehicle turns right and a pedestrian crosses the street perpendicularly, a complex interaction relationship is formed between the pedestrian and the right-turning vehicle. By monitoring the pedestrian's thoracic axial rotational angular velocity and upper limb joint flexion angle in real time, it is possible to determine whether the pedestrian has perceived the right-turning vehicle and taken an evasive action. When a defensive posture of the upper limbs is detected, the system promptly triggers a pedestrian collision warning or automatic emergency braking.

[0151] Industrial applicability This invention provides a three-dimensional kinematic quantitative analysis method and system for pedestrian hazard avoidance behavior in intelligent vehicles. It can be applied to advanced driver assistance systems and active safety systems for intelligent vehicles to identify and predict pedestrian hazard avoidance behavior. By finely marking key human joints and combining their three-dimensional kinematic features, the system can monitor the movement of the pedestrian's lower limbs in real time to identify avoidance behavior. Furthermore, the system incorporates the posture features of the chest and upper limbs to assist in determining avoidance behavior patterns, thereby improving the autonomous driving system's ability to predict potential pedestrian conflict risks. This invention has significant industrial applicability and broad application prospects.

[0152] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. Other modifications can be easily made by those skilled in the art. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and the illustrations shown and described herein.

Claims

1. A three-dimensional kinematic quantitative analysis method for pedestrian hazard avoidance behavior in intelligent vehicles, characterized in that, Includes the following steps: S1: Construct a virtual reality environment for a dangerous scenario of human-vehicle conflict, wherein the virtual reality environment includes at least a vehicle model, a road environment model, and a vehicle traveling according to a preset trajectory and speed; S2: The pedestrian enters the virtual reality environment and performs the crossing task with a crossing gait. Inertial measurement units arranged in multiple parts of the pedestrian's body collect the three-dimensional motion data of the pedestrian's whole body joints in real time during the vehicle approach process. S3: Based on the spatial movement characteristics of pedestrians during the approach of vehicles, pedestrian avoidance behavior is classified into four typical modes: forward avoidance, backward avoidance, avoidance in the direction of vehicle travel, and insufficient reaction time. S4: For each avoidance behavior mode, calculate the three-dimensional kinematic parameters of each joint at the preset analysis time. The three-dimensional kinematic parameters include joint angle, joint angular velocity and joint angular acceleration. S5: Perform quantitative comparative analysis on the three-dimensional kinematic parameters under different avoidance behavior modes, and output the joint three-dimensional kinematic difference characteristics between different avoidance behavior modes, wherein the difference characteristics include at least angular velocity difference and angular acceleration difference.

2. The method according to claim 1, characterized in that, The pedestrian-vehicle conflict scenario described in step S1 is selected from at least one of the following four typical scenarios: the basic scenario of vehicles going straight and pedestrians crossing perpendicularly, the scenario of multiple vehicles driving in parallel and pedestrians crossing perpendicularly with obstructed vision, the scenario of multiple vehicles driving in parallel and pedestrians crossing in the middle with obstructed vision, and the scenario of vehicles turning right and pedestrians crossing perpendicularly with turning interaction.

3. The method according to claim 1, characterized in that, The three-dimensional motion data of the whole body joints mentioned in step S2 includes at least: Motion data of the thorax in the flexion-extension direction, axial rotation direction, and lateral bending direction; Movement data of the shoulder joint in the flexion-extension and abduction-adduction directions; Motion data of the elbow joint in the flexion-extension direction and the ulnar and radial deviation directions; Hip joint motion data in flexion-extension and abduction-adduction directions; And motion data of the knee joint in the flexion-extension direction and the abduction-adduction direction.

4. The method according to claim 1, characterized in that, The classification criteria in step S3 are as follows: when the pedestrian's center of mass displacement or lower limb stride is less than half the length of the pedestrian's stride, it is determined that the pedestrian's reaction is too slow; otherwise, based on the direction of the pedestrian's spatial displacement relative to its initial direction of travel, it is determined to avoid the vehicle by moving forward, avoiding it by moving backward, or avoiding it by moving in the direction of the vehicle's travel.

5. The method according to any one of claims 1 to 4, characterized in that, The method also includes step S6: generating pedestrian behavior recognition parameters based on the three-dimensional kinematic differences in joints among the various avoidance behavior patterns; the behavior recognition parameters include: angular velocity and angular acceleration of the lower limb joints, used to determine the pedestrian's movement tendency during the avoidance process; angular velocity and angular acceleration characteristics of the thoracic cavity in the flexion or extension direction, used to assist in recognizing pedestrian avoidance behavior patterns; flexion angle and adduction angle characteristics of the upper limb joints, used to generate high collision risk warning signals; the behavior recognition parameters are used to input into the advanced driver assistance system or active safety system of the intelligent vehicle.

6. A three-dimensional kinematic quantitative analysis system for pedestrian hazard avoidance behavior in intelligent vehicles, used to implement the method described in any one of claims 1-5, characterized in that, include: The virtual scene construction module is used to construct a virtual reality environment for a pedestrian-vehicle conflict scenario. The virtual reality environment includes at least a vehicle model, a road environment model, and vehicles traveling according to a preset trajectory and speed. The pedestrian-vehicle conflict scenario is selected from at least one of the following four typical scenarios: a basic scenario of vehicles going straight and pedestrians crossing perpendicularly, a scenario of multiple vehicles driving in parallel and pedestrians crossing perpendicularly with obstructed view, a scenario of multiple vehicles driving in parallel and pedestrians crossing in the middle with obstructed view, and a scenario of vehicles turning right and pedestrians crossing perpendicularly. An inertial motion capture module includes inertial measurement units arranged at multiple parts of a pedestrian's body. These units are used to collect real-time three-dimensional motion data of the pedestrian's entire body joints during vehicle approach when the pedestrian is performing a street-crossing task in the virtual reality environment using a street-crossing gait. The three-dimensional motion data of the entire body joints includes at least the following: Motion data of the thorax in the flexion-extension direction, axial rotation direction, and lateral bending direction; Movement data of the shoulder joint in the flexion-extension and abduction-adduction directions; Motion data of the elbow joint in the flexion-extension direction and the ulnar and radial deviation directions; Hip joint motion data in flexion-extension and abduction-adduction directions; And motion data of the knee joint in the flexion-extension direction and the abduction-adduction direction; The behavior pattern classification module is used to classify pedestrian avoidance behavior into four typical patterns based on the spatial movement characteristics of pedestrians during the approach of vehicles: forward avoidance, backward avoidance, avoidance in the direction of vehicle travel, and insufficient reaction time. The kinematic parameter calculation module is used to calculate the three-dimensional kinematic parameters of each joint at a preset analysis time for each avoidance behavior mode. The three-dimensional kinematic parameters include joint angle, joint angular velocity, and joint angular acceleration. The difference feature output module is used to perform quantitative comparative analysis of the three-dimensional kinematic parameters under different avoidance behavior modes, and output the joint three-dimensional kinematic difference features between different avoidance behavior modes. The difference features include at least angular velocity difference and angular acceleration difference.

7. The system according to claim 6, characterized in that, The classification criteria of the behavior pattern classification module are as follows: when the pedestrian's center of mass displacement or lower limb stride is less than half the length of the pedestrian's stride, it is determined that the pedestrian's reaction is too slow; otherwise, based on the spatial displacement direction of the pedestrian relative to its initial direction of travel, it is determined to be forward avoidance, backward avoidance, or avoidance in the direction of vehicle travel.

8. The system according to claim 6, characterized in that, The preset analysis time in the kinematic parameter calculation module is the common moment when the angles of each joint are all within the range of their peak values.

9. The system according to claim 6, characterized in that, The quantitative comparative analysis in the difference feature output module includes: comparing the distribution range and directional characteristics of each joint on the three-dimensional kinematic parameters under different avoidance behavior modes, so as to identify the kinematic differences between active avoidance behavior and inadequate on-the-spot reaction, as well as the kinematic differences between different active avoidance behavior modes.

10. The system according to claim 6, characterized in that, It also includes a parameter generation module for generating pedestrian behavior recognition parameters based on the three-dimensional kinematic differences in joints among the various avoidance behavior patterns. The behavior recognition parameters include: angular velocity and angular acceleration of the lower limb joints, used to determine the pedestrian's movement tendency during the avoidance process; angular velocity and angular acceleration characteristics of the thoracic cavity in the flexion or extension direction, used to assist in recognizing pedestrian avoidance behavior patterns; and flexion and adduction angle characteristics of the upper limb joints, used to generate high collision risk warning signals. The behavior recognition parameters are used to input into the advanced driver assistance system or active safety system of the intelligent vehicle.

Images