Real-time Collision Avoidance Warning Method and System for New Energy Trucks and Pedestrians

Abstract

This application provides a real-time collision avoidance warning method and system for new energy trucks colliding with pedestrians. The method includes: spatiotemporal registration of environmental perception data and vehicle operating status data surrounding the new energy truck to generate near-field environmental representations and vehicle state representations; constructing a dynamic sweep envelope of the vehicle that changes with its motion posture, and generating a new energy vehicle braking reachability domain based on the dynamic braking state in the vehicle operating status data; identifying visible pedestrian targets based on the near-field environmental representation, and generating occluded potential pedestrian targets to obtain a set of pedestrian risk targets; performing multi-moment trajectory prediction on the set of pedestrian risk targets to generate candidate pedestrian trajectories; performing spatiotemporal intersection calculations between the candidate pedestrian trajectories and the vehicle dynamic sweep envelope to determine the collision risk representation; and generating a collision avoidance warning level and warning execution command based on the collision risk representation. This application can improve the timeliness of pedestrian warnings in blind spots, enhance the accuracy of collision risk identification, and improve the adaptability of warning control.

Description

Technical Field

[0001] This application relates to the field of new energy vehicle technology, and in particular to a real-time collision avoidance warning method and system for new energy trucks colliding with pedestrians. Background Technology

[0002] With the rapid development of urban logistics distribution, industrial park transfer, and last-mile transportation, new energy trucks are widely used in urban roads, warehousing parks, residential communities, and enclosed factory areas due to their low noise, low emissions, and low operating costs. However, when starting, turning, reversing, or loading / unloading at low speeds, new energy trucks often share close proximity with pedestrians, non-motorized vehicles, and workers. Blind spots can easily form in front of, below, to the sides and rear of the vehicle, and around the cargo box. Furthermore, the low noise level of new energy vehicles at low speeds makes it relatively difficult for pedestrians to perceive the approaching vehicle.

[0003] Existing pedestrian collision avoidance warning technologies typically identify pedestrians ahead of a vehicle using cameras, radar, or multi-sensor fusion, and generate warning information based on the distance between the pedestrian and the vehicle, relative speed, or collision time threshold. Some technologies further integrate automatic braking or lane keeping control to intervene in forward collision risks. However, these solutions are mostly geared towards ordinary passenger cars or standard forward driving scenarios, with relatively fixed warning areas and risk assessment rules, making it difficult to adapt to the operating characteristics of new energy trucks, which have large vehicle sizes, wide blind spots, complex turning sweep areas, and significant changes in load conditions.

[0004] Especially in situations where pedestrians suddenly enter the vehicle's driving area, trucks make low-speed turns resulting in inner wheel differences, trucks reverse and approach pedestrians, or braking capabilities change due to full load or limited energy recovery, existing technologies struggle to predict the spatiotemporal relationship between pedestrians and vehicles dynamically occupying their areas. They also find it difficult to adjust the warning lead time based on the current braking status of new energy trucks, easily leading to delayed warnings, false alarms, or missed alarms. Therefore, there is an urgent need for a pedestrian collision avoidance warning method that can combine the vehicle's dynamic sweep range, the obstruction of potential pedestrians, and the braking capabilities of new energy vehicles for real-time risk assessment. Summary of the Invention

[0005] In view of this, embodiments of this application provide a real-time collision avoidance warning method and system for pedestrian collisions of new energy trucks, in order to solve the problems of blind spot pedestrians being difficult to predict, insufficient identification of dynamic sweep risk, and poor adaptability of braking state in the existing technology.

[0006] A first aspect of this application provides a real-time collision avoidance warning method for pedestrian collisions involving new energy freight vehicles, comprising: collecting environmental perception data and vehicle operating status data of the new energy freight vehicle, and performing spatiotemporal registration on the environmental perception data and vehicle operating status data to generate near-field environmental representation and vehicle state representation; constructing a vehicle dynamic sweep envelope that changes with the vehicle's motion posture based on the vehicle state representation, and generating a new energy braking reachability domain based on the power braking state in the vehicle operating status data; identifying visible pedestrian targets based on the near-field environmental representation, and generating occluded potential pedestrian targets by combining occlusion boundaries, road traffic constraints, and target motion residues to obtain a set of pedestrian risk targets; performing multi-time trajectory prediction on the set of pedestrian risk targets to generate pedestrian candidate trajectories corresponding to each pedestrian risk target; performing spatiotemporal intersection calculation on the pedestrian candidate trajectories and the vehicle dynamic sweep envelope, and determining a collision risk representation by combining the new energy braking reachability domain; generating a collision avoidance warning level and a warning execution command based on the collision risk representation, and executing driving prompts, external warnings, or driving control limits for the new energy freight vehicle based on the warning execution command.

[0007] A second aspect of this application provides a real-time collision avoidance and warning system for pedestrian collisions involving new energy freight vehicles, comprising: a data acquisition module for acquiring environmental perception data and vehicle operating status data surrounding the new energy freight vehicle, and performing spatiotemporal registration on the environmental perception data and vehicle operating status data to generate near-field environmental representations and vehicle state representations; a construction module for constructing a dynamic sweep envelope of the vehicle that changes with the vehicle's motion posture based on the vehicle state representations, and generating a new energy vehicle braking reachability domain based on the power braking state in the vehicle operating status data; and a generation module for identifying visible pedestrian targets based on the near-field environmental representations and combining... The system combines occlusion boundaries, road traffic constraints, and target motion residues to generate potential pedestrian targets, resulting in a set of pedestrian risk targets. A prediction module performs multi-moment trajectory prediction on the pedestrian risk target set, generating candidate pedestrian trajectories corresponding to each risk target. A calculation module performs spatiotemporal intersection calculations between the candidate pedestrian trajectories and the vehicle dynamic sweep envelope, and determines the collision risk characterization by combining the new energy vehicle braking reachability domain. An execution module generates collision warning levels and warning execution commands based on the collision risk characterization, and executes driving prompts, external warnings, or driving control limits for the new energy truck based on the warning execution commands.

[0008] The above-described technical solutions adopted in the embodiments of this application can achieve the following beneficial effects: By collecting environmental perception data and vehicle operation status data of new energy freight trucks, and performing spatiotemporal registration on these data, near-field environmental representations and vehicle state representations are generated. Based on the vehicle state representation, a dynamic sweep envelope of the vehicle that changes with its motion posture is constructed, and a new energy braking reachability domain is generated based on the dynamic braking status in the vehicle operation status data. Visible pedestrian targets are identified based on the near-field environmental representation, and occluded potential pedestrian targets are generated by combining occlusion boundaries, road traffic constraints, and target motion residues, resulting in a set of pedestrian risk targets. Multi-moment trajectory prediction is performed on the set of pedestrian risk targets to generate candidate pedestrian trajectories corresponding to each risk target. The candidate pedestrian trajectories are spatiotemporally intersected with the vehicle dynamic sweep envelope, and a collision risk representation is determined by combining the new energy braking reachability domain. A collision warning level and warning execution command are generated based on the collision risk representation, and driving prompts, external warnings, or driving control limits are executed on the new energy freight truck based on the warning execution command. This application can improve the timeliness of blind spot pedestrian warnings, enhance the accuracy of collision risk identification, and improve the adaptability of warning control. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart illustrating the real-time collision avoidance and early warning method for pedestrian collisions of new energy freight vehicles provided in this application embodiment; Figure 2 This is a schematic diagram of the structure of the real-time collision avoidance and early warning system for pedestrian collisions of new energy trucks provided in this application embodiment; Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0011] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0012] In existing technologies, pedestrian collision avoidance warnings for new energy trucks typically rely on cameras, radar, or multi-sensor fusion to identify pedestrians in front of the vehicle. They then assess the collision risk based on the relative distance, relative speed, or collision time threshold between the pedestrian and the vehicle, and subsequently issue an alert to the driver. Some solutions also incorporate the vehicle's braking system to perform assisted deceleration or emergency braking. This type of technology can, to a certain extent, identify visible pedestrian targets in front of the vehicle and provide warnings about the risk of pedestrian collisions during forward travel.

[0013] However, in scenarios such as urban delivery, industrial park transfers, warehousing and loading / unloading, and residential community traffic, new energy trucks often operate under complex conditions including low-speed starts, turns, reversing, and driving close to the side of the road. This creates significant blind spots in areas such as the area beneath the front of the vehicle, the sides and rear, around the cargo box, and the inner wheel difference area during turns. Furthermore, new energy trucks generate less noise at low speeds, making it difficult for pedestrians to perceive the approaching vehicle. Additionally, the vehicle's braking performance is affected by factors such as load, dynamic braking status, and energy recovery status. Existing technologies often employ fixed warning areas and fixed risk thresholds, making it difficult to predict potential pedestrians within obstructed areas in a timely manner. They also struggle to accurately identify the spatiotemporal intersection risk between the vehicle's dynamic sweeping area and the pedestrian's trajectory, resulting in problems such as difficulty in predicting pedestrians in blind spots, insufficient identification of dynamic sweeping risks, and poor adaptability to braking states.

[0014] To address the aforementioned issues, this application provides a real-time collision avoidance and early warning method for pedestrian collisions involving new energy freight vehicles. First, it collects environmental perception data and vehicle operating status data surrounding the new energy freight vehicle, and then performs spatiotemporal registration on these two types of data to generate near-field environmental representations and vehicle state representations. This processing method unifies the surrounding targets, occlusion areas, road traffic constraints, and the vehicle's own motion state under the same spatiotemporal reference, providing a consistent data foundation for subsequent risk prediction.

[0015] Furthermore, this application constructs a dynamic sweep envelope of the vehicle that changes with the vehicle's motion posture based on the vehicle state representation, and generates a new energy vehicle braking reachability domain based on the power braking state in the vehicle's operating state data. The dynamic sweep envelope characterizes the actual occupied area formed by the new energy truck during steering, yaw, forward movement, and reversing within a future prediction time window; the new energy vehicle braking reachability domain characterizes the deceleration and stopping boundaries that the vehicle can reach under current power braking conditions. Therefore, this application incorporates the dynamic changes in vehicle outline occupancy and the actual braking capability of the new energy truck into the pedestrian collision risk assessment process.

[0016] Furthermore, this application identifies visible pedestrian targets based on near-field environmental characterization and generates occluded potential pedestrian targets by combining occlusion boundaries, road traffic constraints, and target motion residues, thus obtaining a set of pedestrian risk targets. For pedestrians already identified by sensing devices, this application can predict their trajectories based on their target states; for pedestrians not yet fully visible within occluded areas, this application can establish potential pedestrian targets in advance through the occluded pedestrian potential field, thereby avoiding issuing warnings only after the pedestrian has fully entered the sensor's field of view.

[0017] Furthermore, this application performs multi-time trajectory prediction on the pedestrian risk target set, generating pedestrian candidate trajectories corresponding to each pedestrian risk target. The pedestrian candidate trajectories are then spatiotemporally intersected with the vehicle dynamic sweep envelope, and the collision risk characterization is determined by combining this with the new energy vehicle braking reachability domain. This collision risk characterization not only reflects whether pedestrians and vehicles may spatially overlap, but also further reflects the intersection time, intersection position, spatial margin, braking margin, and risk confidence level, enabling risk assessment to simultaneously consider pedestrian movement trends, vehicle dynamic occupancy range, and the vehicle's current braking capability.

[0018] Based on the aforementioned collision risk characterization, this application generates collision avoidance warning levels and warning execution commands, and executes driving prompts, external warnings, or driving control limits for new energy trucks based on the warning execution commands. Therefore, this application can improve the timeliness of blind spot pedestrian warnings, enhance the accuracy of collision risk identification, and strengthen the adaptability of warning control. It is particularly suitable for real-time collision avoidance warnings for pedestrians in complex scenarios such as low-speed near-field collisions, obstructed intrusions, turning sweeps, reversing approaches, and load changes for new energy trucks.

[0019] The technical solution of this application will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0020] Figure 1 This is a flowchart illustrating the real-time collision avoidance and early warning method for pedestrian collisions involving new energy freight vehicles provided in this application embodiment. Figure 1 As shown, the method may specifically include: S101 collects environmental perception data and vehicle operation status data around new energy trucks, and performs spatiotemporal registration on the environmental perception data and vehicle operation status data to generate near-field environmental representation and vehicle status representation. S102, construct the vehicle dynamic sweep envelope that changes with the vehicle's motion posture based on the vehicle state characterization, and generate the new energy braking reachability domain based on the power braking state in the vehicle operation state data. S103, based on the near-field environment characterization, identify visible pedestrian targets, and combine occlusion boundaries, road traffic constraints and target motion residues to generate occluded potential pedestrian targets, thus obtaining a set of pedestrian risk targets; S104, perform multi-time trajectory prediction on the set of pedestrian risk targets, and generate pedestrian candidate trajectories corresponding to each pedestrian risk target; S105 calculates the spatiotemporal intersection of pedestrian candidate trajectories and vehicle dynamic sweep envelopes, and determines the collision risk characterization by combining the reachability domain of new energy braking. S106 generates a collision warning level and a warning execution command based on the collision risk characterization, and executes driving prompts, external warnings, or driving control limits for new energy trucks based on the warning execution command.

[0021] In some embodiments, environmental perception data and vehicle operating status data of the new energy truck are collected, and spatiotemporal registration of the environmental perception data and vehicle operating status data is performed to generate near-field environmental representation and vehicle status representation, including: The surrounding environment perception data and vehicle operation status data are collected according to a unified time benchmark, and the data collected at different times are time-aligned based on vehicle motion compensation. Based on the vehicle coordinate relationship of new energy trucks, the time-aligned surrounding environment perception data is mapped to a unified vehicle coordinate system to generate near-field occupancy information; The near-field occupancy information and vehicle operating status data are correlated with confidence levels to generate near-field environment representation and vehicle status representation.

[0022] Specifically, when new energy freight vehicles perform urban delivery, park access, or warehousing loading and unloading tasks, the onboard controller synchronously collects environmental perception data and vehicle operating status data according to a unified time reference. This unified time reference can be provided by the system clock of the vehicle controller or by the onboard gateway calibrating the timing of each perception node and vehicle control node. Environmental perception data can originate from visual perception units, radar perception units, and near-field detection units located at the front, sides, rear, and near-field areas of the vehicle body. Vehicle operating status data can originate from the vehicle controller, brake controller, motor controller, and steering controller. During data collection, the system assigns a timestamp to each frame of environmental perception data and each set of vehicle operating status data, ensuring that pedestrian targets, occluded areas, near-field obstacles, and vehicle motion states all have traceable temporal identifiers.

[0023] Because the data sampling frequency, transmission delay, and processing cycle of different sensing units are not entirely consistent, it is necessary to perform time alignment on the data collected at different times before generating the near-field environment representation. Specifically, the current vehicle speed, steering angle, yaw rate, and acceleration in the vehicle's operating status data can be used as the basis for motion compensation. The earlier collected surrounding environment perception data can be extrapolated to their positions, and the later arriving data can be backtracked to ensure that the target positions sensed by each sensor are uniformly corrected to the same target time. For example, when a new energy truck turns right at low speed into the loading and unloading area, the right-side camera identifies the silhouette of a pedestrian near the side of the vehicle at the previous moment, and the millimeter-wave radar returns the near-range reflection point at the next moment. The system compensates for the positions of the pedestrian silhouette and the reflection point based on the changes in the vehicle's steering angle and yaw rate within this time difference, so that the two correspond to the near-field spatial positions under the same vehicle body posture.

[0024] After time alignment, the system performs spatial mapping based on the vehicle coordinates of the new energy truck. The vehicle coordinates include the installation position, orientation, field of view boundaries, and calibration parameters of each sensing unit relative to the vehicle reference point. The system establishes a unified vehicle coordinate system using the vehicle reference point, transforms the time-aligned surrounding environment perception data into this unified coordinate system, and combines this with the vehicle's outer contour boundaries to spatially represent the areas below the front of the vehicle, the lateral blind spots, the rear of the vehicle, and the area around the cargo box. For visual perception data, pedestrian frames, human keypoints, passable boundaries, and occlusion boundaries can be mapped to target areas in the vehicle coordinate system; for radar perception data, distance, orientation, and speed information can be converted into target occupancy points; for near-range detection data, short-range obstacle states can be filled into the corresponding near-field grid areas. The resulting near-field occupancy information can represent the visible target areas, occluded areas, empty areas, and uncertain areas around the vehicle.

[0025] Furthermore, the system correlates near-field occupancy information with vehicle operating status data based on confidence levels. Specifically, environmental confidence levels are assigned to target areas in the near-field occupancy information based on the stability of the sensing source, the number of consecutive target observations, the consistency of target motion, the sensor overlap verification status, and the magnitude of vehicle attitude changes. Simultaneously, state confidence levels are assigned to vehicle operating status data based on the sampling integrity of vehicle status data, the consistency of control node feedback, and the smoothness of operating status changes. Subsequently, the system performs correlation calculations between environmental and state confidence levels, eliminating abnormal sensing points that clearly do not conform to vehicle motion constraints, and retaining target occupancy areas that match the vehicle's current driving direction, steering trend, and near-field spatial structure. For example, when the right front corner camera detects a suspected pedestrian target, and the right-side radar and near-field detection unit both have corresponding occupancy responses in adjacent timeframes, the target area is assigned a high confidence level; when a single sensor generates a short-term isolated target that is discontinuous with the vehicle's motion-compensated spatial position, the target area is assigned a low confidence level.

[0026] Through the above processing, the system generates near-field environment representations and vehicle state representations. The near-field environment representation includes the target's location, occlusion boundaries, passable areas, uncertain areas, and corresponding confidence levels in a unified vehicle coordinate system. The vehicle state representation includes vehicle speed, attitude change trends, steering status, braking response status, and operational state confidence levels. These near-field environment and vehicle state representations provide a unified data foundation for subsequent vehicle dynamic sweep envelope construction, potential pedestrian generation, pedestrian trajectory prediction, and collision risk assessment, thereby improving the temporal consistency and spatial accuracy of near-field pedestrian risk identification.

[0027] In some embodiments, a vehicle dynamic sweep envelope that varies with the vehicle's motion posture is constructed based on the vehicle state representation, and a new energy braking reachability domain is generated based on the power braking state in the vehicle operating state data, including: The attitude evolution trajectory of the new energy truck within the prediction time window is determined based on the vehicle state characterization, and the vehicle outline constraint is projected along the attitude evolution trajectory in a time sequence to generate a vehicle dynamic sweep envelope covering the changes in the vehicle's near-field occupancy. Based on the dynamic braking state, energy recovery constraints, mechanical braking constraints, and load constraints that affect the vehicle's deceleration response are extracted. The deceleration boundary and stopping boundary within the prediction time window are constructed to generate the new energy braking reachable domain.

[0028] Specifically, when new energy trucks are in operating scenarios such as low-speed start-up, turning, reversing to a platform, or loading / unloading at the side, the system first extracts the current vehicle speed, steering angle, yaw rate, driving direction, acceleration, and vehicle outline reference information based on the vehicle state representation, and then recursively predicts the subsequent motion state of the vehicle according to a preset prediction time window. The prediction time window can be dynamically determined based on the current vehicle speed and near-field risk density. For example, when turning right at low speed into the loading / unloading position in a warehouse park, the system continuously calculates the position changes of the front of the truck, the side of the vehicle body, and the rear edge of the cargo box at each sampling time with a short sampling interval, forming a vehicle attitude evolution trajectory. This attitude evolution trajectory does not only represent the movement path of the vehicle's center point, but also includes changes in vehicle steering, yaw rate, and vehicle orientation, so that the subsequent swept area can reflect the continuous changes in the actual space occupied by the new energy truck.

[0029] After determining the attitude evolution trajectory, the system projects the vehicle outline constraints along this trajectory in a temporal sequence. These constraints include the leading edge of the vehicle front, the sides of the vehicle body, the cargo box boundary, the contour safety margin, and the near-field perception blind zone boundary. At each sampling moment, the system maps the vehicle outline constraints to the vehicle position under the corresponding attitude node and expands the connectivity of the outline boundary changes between adjacent sampling moments, generating a continuous near-field occupancy region for the vehicle. For turning scenarios, the system includes the vehicle front outward swing area, the vehicle side sweep area, and the inner wheel difference coverage area in the same envelope; for reversing scenarios, the system includes the vehicle rearward sweep area and the rear blind zone expansion area in the same envelope. The resulting dynamic vehicle sweep envelope can express the spatial range that the vehicle may occupy within the prediction time window, providing a dynamic spatial reference for pedestrian trajectory intersection judgment.

[0030] Simultaneously, the system generates a new energy braking reachability domain based on the dynamic braking status in the vehicle's operating status data. Specifically, it extracts energy recovery constraints, mechanical braking constraints, and load constraints from the dynamic braking status. The energy recovery constraint characterizes the motor's anti-drag capability and its available deceleration capability after being limited by battery status; the mechanical braking constraint characterizes the hydraulic braking response and braking pressure build-up process; and the load constraint characterizes the impact of cargo weight and axle load changes on braking distance. The system correlates these constraints with the current vehicle speed, road surface adhesion estimation, and braking response delay to calculate the deceleration boundary within the prediction time window and further extrapolates the stopping boundaries that the vehicle can reach under different risk levels.

[0031] For example, when a fully loaded new energy truck enters a residential intersection with its battery in a state of limited energy recovery, the system reduces the weight of the available deceleration corresponding to the energy recovery constraint and increases the impact of mechanical braking constraints and load constraints on the stopping boundary, thus extending the new energy vehicle's braking reachability domain forward compared to the unloaded state. When the vehicle is unloaded and the braking response is stable, the system forms a shorter stopping boundary based on the higher available deceleration. In this way, the new energy vehicle's braking reachability domain can reflect the vehicle's current actual deceleration capability in real time, rather than using a fixed braking distance or a fixed warning threshold.

[0032] By constructing a vehicle dynamic sweep envelope and a new energy braking reachability domain, this embodiment integrates the future near-field occupancy changes of new energy trucks with the current power braking capability into the risk judgment basis, so that the spatiotemporal intersection calculation between the subsequent pedestrian candidate trajectory and the vehicle movement area has more accurate spatial constraints and braking constraints, thereby improving the timeliness and adaptability of pedestrian collision warning in scenarios such as turning, reversing, full load and limited energy recovery.

[0033] In some embodiments, the vehicle outline constraints are temporally projected along the attitude evolution trajectory to generate a vehicle dynamic sweep envelope covering changes in the vehicle's near-field occupancy, including: The vehicle body outline constraints are converted into a vehicle body boundary model that varies with the vehicle's attitude. Based on the sampling time within the prediction time window, the vehicle boundary model is mapped to the pose node corresponding to the attitude evolution trajectory to generate the temporal occupied region; Based on the continuous change relationship of vehicle body boundaries between adjacent pose nodes, the temporal occupied region is expanded and envelope merged to obtain the vehicle dynamic sweep envelope.

[0034] Specifically, after the new energy truck obtains its attitude evolution trajectory within the predicted time window based on the vehicle state characterization, the system first models the vehicle body outline constraints. These constraints can be determined based on the vehicle's measured dimensions, cargo box boundaries, front and rear bumper boundaries, wheel envelope boundaries, and preset safety margins. A vehicle body boundary model is then established using the vehicle's center of mass or rear axle center as a reference. This boundary model is not a static rectangle fixed in map coordinates but is bound to the vehicle's heading angle, steering angle, yaw rate, and driving direction, allowing it to update synchronously with changes in vehicle attitude. For new energy vans, the front edge of the cab, the side edge of the cab, the side edge of the cargo box, the rear edge, and the near-ground blind spot edge can be uniformly represented as closed boundaries, with extended boundaries related to low-speed near-field risks set at the front side and rear rear of the vehicle body.

[0035] During temporal projection, the system reads pose nodes from the attitude evolution trajectory according to the sampling time within the prediction time window. Each pose node includes the vehicle position, heading angle, yaw state, and driving direction. The system maps the vehicle boundary model to the unified vehicle coordinate system or local road coordinate system corresponding to each pose node, obtaining the vehicle occupancy area at each sampling time. Taking a vehicle turning right into a community delivery lane as an example, at the current moment, the front of the vehicle is not yet close to the edge of the sidewalk, but in subsequent pose nodes, the leading edge of the front of the vehicle will swing outward, and the inner side of the cargo box will sweep towards the curb. The system calculates the occupancy boundaries corresponding to the front of the vehicle, the body, and the cargo box at each sampling time, thus forming a temporal occupancy area arranged over time.

[0036] Since there is still a continuous motion process between adjacent pose nodes, if only the occupied area at discrete sampling moments is retained, the space traversed by the vehicle outline between two sampling moments may be missed. Therefore, the system expands the temporal occupied area based on the continuous change relationship of the vehicle boundary between adjacent pose nodes. Specifically, based on the translation, rotation, and boundary correspondence of adjacent pose nodes, the transition sweep area of ​​the vehicle boundary between nodes can be determined, and the transition sweep area is merged with the vehicle occupied area corresponding to the two nodes. For turning, the system focuses on expanding the coverage areas of the vehicle front swing, lateral sweep, and inner wheel difference; for reversing, the system focuses on expanding the coverage areas of the rear boundary and rear blind spot; for starting from the side of the road, the system focuses on expanding the near-field angle area between the vehicle side and the curb.

[0037] During envelope merging, the system spatially fuses the vehicle-occupied areas and transition sweep areas between adjacent nodes at each sampling time, removing duplicate regions and preserving continuous boundaries to obtain a dynamic vehicle sweep envelope covering the near-field occupancy changes within the prediction time window. Simultaneously, the system can label the dynamic vehicle sweep envelope according to time and orientation attributes, enabling subsequent pedestrian candidate trajectories to perform spatiotemporal intersection calculations with vehicle-occupied areas at different time periods and orientations. For example, when a pedestrian is located behind a vehicle on the right side, obscuring its path, the system can determine whether the pedestrian will enter the cargo box's lateral sweep area in the near future, rather than simply determining whether the pedestrian is currently within the vehicle's outline.

[0038] Through the above processing, this embodiment transforms the vehicle outline, attitude evolution trajectory, and continuous motion sweep process of the new energy truck into a vehicle dynamic sweep envelope that can be used for risk prediction. This allows for the continuous expression of near-field occupancy changes under conditions such as vehicle turning, reversing, and starting close to the edge, thereby improving the spatial integrity of subsequent pedestrian trajectory intersection recognition and the accuracy of collision warning.

[0039] In some embodiments, visible pedestrian targets are identified based on near-field environmental characterization, and occluded potential pedestrian targets are generated by combining occlusion boundaries, road traffic constraints, and target motion residues, resulting in a set of pedestrian risk targets, including: Pedestrian explicit identification and motion consistency verification are performed on the near-field environment characterization to determine visible pedestrian targets; The potential field of occluded pedestrians is constructed based on the occlusion boundary and road traffic constraints, and the potential field of occluded pedestrians is extended in a short time using the target motion residue to generate occluded potential pedestrian targets with occurrence position constraints and motion direction constraints. Visible pedestrian targets and occluded potential pedestrian targets are fused according to target confidence levels to obtain a pedestrian risk target set.

[0040] Specifically, when new energy trucks are traveling on park roads, in residential communities, or in loading and unloading lanes, the system performs explicit pedestrian identification in the space surrounding the vehicle based on the aforementioned near-field environmental characterization. The near-field environmental characterization includes the target's occupancy position, motion state, occlusion boundary, and target confidence in a unified vehicle coordinate system. The system first extracts target regions with pedestrian morphological characteristics, motion continuity, and near-field occupancy characteristics from these regions, and then performs motion consistency verification by combining the target position changes at adjacent time points.

[0041] For pedestrian silhouettes identified by the visual perception unit, the system can perform cross-validation by combining radar reflection points, close-range detection results, and target movement direction. For suspected human targets in nighttime or low-light scenarios, the system can confirm them by combining heat source distribution and low-speed movement characteristics. If the target's position changes, velocity direction, and spatial occupancy relationships within consecutive time intervals all conform to pedestrian motion constraints, it is identified as a visible pedestrian target, and the corresponding current position, movement direction, target confidence level, and observation continuity are recorded.

[0042] While identifying visible pedestrian targets, the system further constructs an occlusion potential field based on occlusion boundaries and road traffic constraints in the near-field environment representation. Occlusion boundaries can be formed by parked vehicles, cargo boxes, corners, shelves, roadside facilities, and the edge of the vehicle's blind spot. Road traffic constraints can be determined based on pedestrian walkways, loading / unloading area boundaries, curbs, intersections, and passable spaces. The system uses the side of the occlusion boundary closest to the passable area as the potential occurrence area and combines this with the spatial channels through which pedestrians can enter the vehicle's near field to generate the occlusion potential field. This occlusion potential field represents the distribution of potential pedestrians who are not fully visible in the current perception field but may enter the vehicle's dynamic sweep range from the occlusion area.

[0043] Furthermore, the system utilizes the target motion remnant to perform a short-term extension of the occluded pedestrian potential field. The target motion remnant can originate from pedestrian targets that disappear briefly, the direction of movement before occlusion, local occupancy changes in consecutive frames, and motion responses near the occlusion edge. For example, in a warehouse loading and unloading area, a pedestrian passes by the side of a shelf and is then occluded by the cargo box of a van. Based on the pedestrian's direction of movement, speed range, and occlusion boundary position before disappearing, the system extends the occluded pedestrian potential field in the direction in which the pedestrian might continue to move, forming occluded potential pedestrian targets with constraints on appearance location and direction of movement. For occluded areas that are not directly identified, the system can also generate low-confidence occluded potential pedestrian targets based on road traffic constraints and the opening direction of the occlusion boundary.

[0044] After generating visible pedestrian targets and occluded potential pedestrian targets, the system fuses them according to target confidence levels to obtain a set of pedestrian risk targets. During fusion, the system associates targets that are spatially close, move in the same direction, and are temporally continuous to avoid the same pedestrian being repeatedly represented. For visible pedestrian targets, the target confidence level is mainly determined by the explicit recognition results, motion consistency, and multi-source perception verification status. For occluded potential pedestrian targets, the target confidence level is mainly determined by the credibility of the occlusion boundary, the matching degree of road traffic constraints, and the extension intensity of the target motion residual. The fused set of pedestrian risk targets includes both visible pedestrians and potential pedestrians who may enter the near field of vehicles from occluded areas, and each pedestrian risk target is configured with target type, position constraints, motion direction constraints, and target confidence level.

[0045] Through the above processing, this embodiment can not only identify pedestrian targets within the current visible range of the sensor, but also generate potential pedestrian targets in advance based on occlusion boundaries, road traffic constraints, and target motion traces. This enables the pedestrian risk target set to cover both explicit and implicit risks, improving the pedestrian prediction capability and timely warning of new energy trucks in scenarios such as occlusion intrusion, low-speed turning, and near-field blind spots.

[0046] In some embodiments, multi-time trajectory prediction is performed on the pedestrian risk target set to generate pedestrian candidate trajectories corresponding to each pedestrian risk target, including: Based on the target state, spatial constraints, and target confidence of each pedestrian risk target in the pedestrian risk target set, construct the corresponding motion intention representation; By correlating motion intention representations with near-field environment representations, the passable areas and motion boundaries of each pedestrian risk target within the prediction time window can be determined. Based on the passable area and movement boundary, multi-time trajectory extrapolation is performed to generate candidate pedestrian trajectories corresponding to each pedestrian risk target.

[0047] Specifically, after obtaining the set of pedestrian risk targets, the system establishes a target state for trajectory prediction for each pedestrian risk target in the set. The target state can include the current position, movement speed, movement direction, orientation, and target confidence. For occluded potential pedestrian targets, the target state also includes the possible location, possible direction, and confidence decay state determined by the occlusion boundary and the target's motion residual. The system associates the target state with near-field spatial constraints to form a corresponding motion intention representation. This motion intention representation is used to describe the pedestrian risk target's tendency to continue straight, cross, approach the vehicle side, go around the vehicle's front, or rush in from the occlusion boundary within the prediction time window, rather than simply extrapolating linearly based on the current instantaneous speed.

[0048] When constructing motion intent representations, the system determines the positional relationship between pedestrian risk targets and the vehicle's dynamic sweep envelope, road boundaries, and occlusion boundaries. For example, when a new energy truck makes a low-speed right turn on a residential road, if a pedestrian is visible to the right front at the pedestrian crossing exit and facing the vehicle's path, the system configures this as a motion intent representation with a tendency to cross. If there is a potential pedestrian target behind the vehicle, obscured by a corner of a wall, and this obscuration boundary connects to the pedestrian crossing, the system configures this as a motion intent representation with a tendency to intrude into the obscuration. For visible pedestrians with high target confidence, the system increases the weight of the current motion trend; for occluded potential pedestrians approaching the vehicle's sweep area, the system retains multiple possible motion directions.

[0049] Furthermore, the system correlates motion intention representation with near-field environment representation to determine the passable area and motion boundary of pedestrian risk targets within the prediction time window. The passable area can be determined by road access areas, pedestrian walkways, loading / unloading aisles, and open areas around vehicles. The motion boundary can be jointly defined by curbs, walls, parked vehicles, shelves, cargo box edges, and non-passable occupied areas. The system performs constraint matching under a unified vehicle coordinate system, eliminating motion directions that clearly cross non-passable areas and retaining candidate motion ranges consistent with pedestrian motion intention, the direction of obstructed openings, and the near-field spatial structure. In the warehouse loading / unloading area, if a pedestrian is located in a narrow passage between the shelf and the vehicle side, the system restricts the passable area to the direction of the passage's extension and its exit area.

[0050] Subsequently, the system performs multi-moment trajectory extrapolation based on the passable area and movement boundaries. Starting with the current or potential location of the pedestrian risk target, the system recursively extrapolates the reachable location according to multiple sampling moments within the prediction time window, and generates multiple trajectory branches based on the range of movement speed, the range of direction changes, spatial boundary constraints, and target confidence. For visible pedestrian targets, the trajectory branches unfold around their current movement trend and orientation; for occluded potential pedestrian targets, the trajectory branches unfold around the opening of the occlusion boundary, the potential location, and the passable space. The system assigns temporal confidence to each trajectory branch and merges trajectory branches that are spatially close, have the same direction, and the same risk meaning to generate candidate pedestrian trajectories corresponding to each pedestrian risk target.

[0051] Through the above processing, this embodiment incorporates pedestrian target state, spatial traffic constraints, target confidence, and near-field environment structure into the trajectory prediction process, enabling the generated pedestrian candidate trajectory to cover the movement trend of visible pedestrians and the possibility of occlusion of potential pedestrians, thereby improving the completeness of subsequent spatiotemporal intersection calculations and the advance and accuracy of pedestrian collision warnings for new energy trucks.

[0052] In some embodiments, multi-moment trajectory extrapolation is performed based on traversable areas and movement boundaries to generate candidate pedestrian trajectories corresponding to each pedestrian risk target, including: Taking the current location of the pedestrian risk target as the trajectory starting point, the sequence of reachable locations is determined based on the passable area, and the state transition between adjacent reachable locations is constrained by the motion boundary. The state transition is probabilistically weighted according to the target confidence level, forming multiple predicted trajectory branches; The predicted trajectory branches are filtered by time series confidence and the trajectories are merged to generate candidate trajectories for pedestrians.

[0053] Specifically, after determining the passable area and movement boundary of each pedestrian risk target, the system uses the current position of the pedestrian risk target as the trajectory starting point and extrapolates the possible locations it may reach within the prediction time window. For visible pedestrian targets, the current position can be determined by the target's center point, foot landing area, or body-occupied area in the continuous perception results; for occluded potential pedestrian targets, the current position can be determined by the potential appearance location near the occlusion boundary.

[0054] Based on the spatial connectivity of the passable area, the system expands the sequence of reachable locations outward from the trajectory starting point, ensuring that the sequence of reachable locations always lies within the passable area. For example, when a new energy truck makes a low-speed right turn on a park road, and a pedestrian is located at the pedestrian walkway exit on its right front side, the system uses the pedestrian's current position at that exit as the trajectory starting point and generates multiple reachable locations along the pedestrian walkway extension direction, the direction of crossing the lane, and the direction closer to the vehicle side.

[0055] During the generation of reachable location sequences, the system utilizes motion boundaries to constrain state transitions between adjacent reachable locations. Motion boundaries can include curb boundaries, wall boundaries, parked vehicle boundaries, shelf boundaries, occlusion boundaries, and areas impassable to vehicles in the near field. The system determines whether the line connecting adjacent reachable locations crosses impassable areas and, based on pedestrian speed variation range, direction variation range, and short-term stopping variation range, eliminates or downgrades state transitions that do not conform to motion boundary constraints. For example, in a warehouse loading / unloading area, if a pedestrian is located in a narrow passage between the shelf and the side of a truck, the system restricts their state transitions to unfold along the passage direction and the passage exit direction, avoiding the generation of invalid trajectories that cross shelf or cargo container areas.

[0056] Furthermore, the system assigns probabilistic weights to state transitions based on target confidence. For visible pedestrian targets with high confidence and stable continuous observation, the system increases the state transition probability consistent with the current direction of movement, orientation, and historical movement trend. For potential pedestrian targets with low confidence or those generated by occlusion of the pedestrian potential field, the system retains multiple possible entry directions and assigns state transition probabilities based on the confidence of the occlusion boundary, the degree of matching of road traffic constraints, and the extension intensity of the target's motion residual. For example, if a pedestrian has a residual trace of movement towards the lane before being occluded by a parked vehicle, the system increases the state transition probability of the pedestrian entering the near-field region of the vehicle from the occlusion boundary. If a potential target is inferred solely from a static occlusion boundary, the corresponding trajectory branch maintains a low confidence level but is not directly deleted.

[0057] Furthermore, after completing multiple rounds of state transitions, the system forms multiple predicted trajectory branches. Each predicted trajectory branch includes reachable location, direction of movement, trajectory probability, and temporal confidence at multiple prediction times. The system filters the predicted trajectory branches based on temporal confidence, deleting trajectory branches with confidence levels below preset conditions for consecutive times and those that do not match near-field environmental constraints. Trajectory branches with similar spatial locations, consistent directions of movement, and the same risk implications are merged. During merging, the system retains the central trajectory that represents the type of movement trend and inherits the comprehensive confidence of each branch before merging, ensuring that the pedestrian candidate trajectories cover the main movement possibilities while avoiding an excessive number of trajectories that could affect real-time calculations.

[0058] Through the above processing, this embodiment can combine the passable area, motion boundary and target confidence into the multi-time trajectory extrapolation process, making the generated pedestrian candidate trajectory more consistent with the near-field road structure of new energy trucks and the actual motion constraints of pedestrians. This improves the accuracy and real-time performance of subsequent spatiotemporal intersection calculations of pedestrian candidate trajectories and vehicle dynamic sweep envelopes, and reduces false alarms caused by invalid trajectories.

[0059] In some embodiments, the candidate pedestrian trajectory and the vehicle dynamic sweep envelope are spatiotemporally intersected, and the collision risk characterization is determined by combining the new energy vehicle braking reachability domain, including: Based on the predicted time window, the candidate trajectories of pedestrians and the dynamic sweep envelope of vehicles are matched in time sequence to determine the intersection status of pedestrian risk targets entering the vehicle-occupied area. Calculate the corresponding rendezvous time, rendezvous position, and spatial margin based on the rendezvous status; By coupling the meeting time, meeting location, and spatial margin with the reachability domain of the new energy vehicle for evaluation, a collision risk characterization including braking margin and risk confidence level is generated.

[0060] Specifically, after obtaining the pedestrian candidate trajectories and vehicle dynamic sweep envelopes corresponding to each pedestrian risk target, the system performs temporal matching between the two according to the same prediction time window. The pedestrian candidate trajectories include the pedestrian reachable locations and trajectory confidence at each prediction time, while the vehicle dynamic sweep envelopes include the vehicle-occupied areas and continuous sweep areas at each prediction time. Using the same sampling time as an index, the system spatially superimposes the pedestrian reachable locations, pedestrian-occupied areas, and vehicle-occupied areas, and combines the continuous motion relationship between adjacent sampling times to determine whether the pedestrian risk target has entered the vehicle-occupied area, as well as the order and duration of entry into the vehicle-occupied area. For occluded potential pedestrian targets, the system also performs weighted confirmation of the intersection state based on their trajectory confidence and the confidence decay state of the occluded pedestrian potential field, avoiding directly treating low-confidence potential trajectories as high-risk events.

[0061] After determining the meeting state, the system further calculates the meeting time, meeting position, and spatial margin corresponding to the meeting state. The meeting time is used to indicate the predicted time when a pedestrian risk target first enters the vehicle's dynamic sweep envelope or reaches the minimum distance with the vehicle's sweep boundary; the meeting position is used to indicate the position of the pedestrian risk target in the unified vehicle coordinate system when it overlaps or approaches the vehicle's dynamic sweep envelope; the spatial margin is used to indicate the minimum lateral distance, longitudinal distance, or comprehensive safety distance between the pedestrian's occupied area and the vehicle's sweep boundary. For example, when a new energy truck turns right into the loading and unloading area, the system can determine whether a potential pedestrian at the right-side obstruction boundary will enter the cargo box's lateral sweep area in 1.2 seconds, and determine that the meeting position is located in the near-field area on the right rear side of the vehicle, with a spatial margin less than the safety distance at the current vehicle speed.

[0062] Subsequently, the system couples the meeting time, meeting position, and spatial margin with the new energy braking reachability domain for evaluation. The new energy braking reachability domain includes the deceleration boundary and stopping boundary corresponding to the vehicle under the current energy recovery constraint, mechanical braking constraint, and load constraint. The system compares the meeting time with the braking response delay, performs a spatial comparison between the meeting position and the stopping boundary, and matches the spatial margin with the vehicle's reachability range under different deceleration boundaries to determine whether the vehicle has sufficient braking margin under the current dynamic braking state. If the vehicle is fully loaded, in a state of limited energy recovery, or in a state of slow mechanical braking response, the system will reduce the available braking margin and increase the base value of the risk level corresponding to the same meeting state; if the vehicle is in a state of low-speed unloaded and stable braking response, the system will adjust the risk level accordingly based on the shorter stopping boundary.

[0063] Furthermore, the system combines the trajectory confidence of pedestrian candidate trajectories, the attitude prediction confidence of the vehicle's dynamic sweep envelope, and the state confidence of the new energy vehicle's braking reachability domain to generate a risk confidence score. It also encodes the urgency of the encounter, the trend of spatial margin changes, and the braking margin into a unified collision risk characterization. This collision risk characterization can be used to subsequently determine the collision warning level and the warning execution command. For example, when the visible pedestrian trajectory overlaps with the vehicle's front sweep area within a short period and the current braking margin is insufficient, the system generates a high-risk collision risk characterization; when a potential pedestrian is obscured and can only approach the side of the vehicle and the braking margin is sufficient, the system generates a medium-to-low-risk collision risk characterization.

[0064] Through the above processing, this embodiment unifies and couples the pedestrian candidate trajectory, the vehicle dynamic sweep envelope, and the new energy vehicle braking reachability domain, so that the collision risk judgment simultaneously reflects the spatiotemporal relationship of pedestrians entering the vehicle-occupied area and the current braking capability of the new energy vehicle truck, thereby improving the accuracy of collision risk identification and enhancing the early warning adaptability under scenarios of full load and changing braking capability.

[0065] In some embodiments, a collision warning level and a warning execution command are generated based on the collision risk characterization, and driving prompts, external warnings, or driving control limits for the new energy truck are executed based on the warning execution command, including: Based on the risk confidence level, meeting urgency and braking margin in the collision risk characterization, determine the collision warning level corresponding to the pedestrian risk target; Based on the collision avoidance warning level, target orientation relationship, and control executable boundary, generate collaborative warning execution instructions; According to the collaborative early warning execution command, the driver's terminal prompt status, the vehicle external directional reminder status, and the vehicle driving output boundary are linked for control.

[0066] Specifically, after obtaining the collision risk characterization, the system first extracts the risk confidence level, meeting urgency level, and braking margin from the collision risk characterization, and uses these three as the main basis for determining the collision warning level. The risk confidence level is used to represent the credibility of the pedestrian risk target and its candidate trajectory; the meeting urgency level is used to represent the time urgency of the pedestrian risk target entering the vehicle's dynamic sweep envelope; and the braking margin is used to represent the deceleration space of the new energy truck relative to the meeting position under the current dynamic braking state.

[0067] The system jointly assesses the above information. When the risk confidence level is high, the meeting point is close, and the braking margin is insufficient, a higher collision warning level is determined. When the risk confidence level is low or the meeting point is still at the edge of the vehicle's dynamic sweep envelope, a lower collision warning level is determined. Thus, the collision warning level can simultaneously reflect the authenticity of the pedestrian risk target, the urgency of the collision, and the vehicle's current ability to take evasive action.

[0068] After determining the collision warning level, the system further generates coordinated warning execution commands based on the collision warning level, target orientation relationship, and controllable execution boundaries. The target orientation relationship can be determined based on the position of the pedestrian risk target relative to the front, side, rear, and turning sweep area of ​​the vehicle. The controllable execution boundaries can be determined based on the current vehicle speed, driving status, braking response status, and driver operation status. For pedestrian risk targets located in front of the vehicle with a high degree of urgency, the system generates a strong warning command for the driver and a limiting command for the vehicle's driving output boundary. For pedestrian risk targets located in the sweep area to the side or right rear of the vehicle, the system generates an external warning command with directional orientation. For potential pedestrian targets obstructed by the pedestrian potential field, the system can generate an obstruction area warning command and a low-speed torque limiting command, so that the vehicle enters a controlled warning state before the pedestrian is fully visible.

[0069] Furthermore, the system coordinates the control of the driver's alert status, external directional warning status, and vehicle driving output boundary according to the collaborative warning execution command. The driver's alert status can be expressed through instrument display, sound prompts, seat vibration, or directional graphic prompts, highlighting the area where the risk target is located based on the target's orientation. The external directional warning status can control the operation of the sound and light warning units in the corresponding positions of the vehicle body, so that pedestrians near the vehicle's blind spot or obstruction boundary receive a vehicle approach warning. The vehicle driving output boundary can be constrained by limiting drive torque, reducing creep speed, increasing brake pre-charge pressure, or enhancing available deceleration response.

[0070] For example, when a new energy truck makes a low-speed right turn in a residential area, if the trajectory of a potential pedestrian at the right front obstruction boundary will enter the side sweep area of ​​the cargo box in a short time, and the current vehicle is fully loaded, resulting in insufficient braking margin, the system will simultaneously output a right-side risk warning to the driver, control the right-side external sound and light warning unit to work, and limit the vehicle's drive output.

[0071] During the execution of coordinated control, the system can also record feedback information such as driver braking, steering, releasing the accelerator pedal, pedestrian avoidance, and target disappearance. This feedback information is then correlated with the collision avoidance warning level and the coordinated warning execution command for subsequent updates to scenario-based warning parameters. For example, if the driver brakes immediately after a medium-level warning and the pedestrian risk target is far from the vehicle's dynamic sweep envelope, the system can retain the warning parameters corresponding to that scenario. If insufficient braking margin still occurs after a high-level warning, the system can increase the warning lead time under similar load and orientation scenarios.

[0072] Through the above processing, this embodiment can generate a collision warning level and a coordinated warning execution command that matches the risk level, target location and vehicle control capability based on the collision risk characterization, and realize the linkage response of the driver end, the external warning end and the vehicle control end, thereby improving the timeliness, directionality and control adaptability of near-field pedestrian warning for new energy trucks.

[0073] The following are system embodiments of this application, which can be used to execute the method embodiments of this application. For details not disclosed in the system embodiments of this application, please refer to the method embodiments of this application.

[0074] Figure 2 This is a schematic diagram of the structure of the real-time collision avoidance and early warning system for pedestrian collisions of new energy freight vehicles provided in this application embodiment. Figure 2 As shown, the system includes: The acquisition module 201 is used to acquire environmental perception data and vehicle operation status data around the new energy truck, and to perform spatiotemporal registration on the environmental perception data and vehicle operation status data to generate near-field environmental representation and vehicle status representation. The construction module 202 is used to construct a vehicle dynamic sweep envelope that changes with the vehicle's motion posture based on the vehicle state characterization, and to generate a new energy braking reachable domain based on the power braking state in the vehicle operating state data. The generation module 203 is used to identify visible pedestrian targets based on near-field environmental characterization, and generate occluded potential pedestrian targets by combining occlusion boundaries, road traffic constraints and target motion residues, thus obtaining a set of pedestrian risk targets; Prediction module 204 is used to predict the trajectory of pedestrian risk targets at multiple time points and generate candidate trajectories of pedestrians corresponding to each pedestrian risk target. The calculation module 205 is used to perform spatiotemporal intersection calculation of pedestrian candidate trajectories and vehicle dynamic sweep envelopes, and combine the new energy braking reachability domain to determine the collision risk characterization. The execution module 206 is used to generate a collision warning level and a warning execution command based on the collision risk characterization, and to execute driving prompts, external warnings or driving control limits for new energy trucks based on the warning execution command.

[0075] In some embodiments, Figure 2 The acquisition module 201 collects ambient environment perception data and vehicle operation status data according to a unified time reference, and performs time alignment on data collected at different times based on vehicle motion compensation; according to the vehicle coordinate relationship of the new energy truck, the time-aligned ambient environment perception data is mapped to a unified vehicle coordinate system to generate near-field occupancy information; the near-field occupancy information and vehicle operation status data are correlated with confidence to generate near-field environment representation and vehicle status representation.

[0076] In some embodiments, Figure 2 The construction module 202 determines the attitude evolution trajectory of the new energy truck within the prediction time window based on the vehicle state characterization, and projects the vehicle outline constraints along the attitude evolution trajectory in a time sequence to generate a vehicle dynamic sweep envelope covering the changes in the vehicle's near-field occupancy; based on the power braking state, it extracts the energy recovery constraints, mechanical braking constraints, and load constraints that affect the vehicle's deceleration response, constructs the deceleration boundary and parking boundary within the prediction time window, and generates the new energy braking reachable domain.

[0077] In some embodiments, Figure 2 The construction module 202 converts the vehicle body outline constraints into a vehicle body boundary model that changes with the vehicle attitude; according to the sampling time within the prediction time window, the vehicle body boundary model is mapped to the pose nodes corresponding to the attitude evolution trajectory to generate the temporal occupied region; based on the continuous change relationship of the vehicle body boundary between adjacent pose nodes, the temporal occupied region is connected, expanded and envelope merged to obtain the vehicle dynamic sweep envelope.

[0078] In some embodiments, Figure 2The generation module 203 performs explicit pedestrian identification and motion consistency verification on the near-field environment representation to determine visible pedestrian targets; it constructs an occluded pedestrian potential field based on occlusion boundaries and road traffic constraints, and uses the target motion residue to perform short-term extension on the occluded pedestrian potential field to generate occluded potential pedestrian targets with occurrence location constraints and motion direction constraints; it then fuses the visible pedestrian targets and occluded potential pedestrian targets according to target confidence to obtain a pedestrian risk target set.

[0079] In some embodiments, Figure 2 The prediction module 204 constructs a corresponding motion intention representation based on the target state, spatial constraints, and target confidence of each pedestrian risk target in the pedestrian risk target set; it associates the motion intention representation with the near-field environment representation to determine the passable area and motion boundary of each pedestrian risk target within the prediction time window; and it performs multi-time trajectory extrapolation based on the passable area and motion boundary to generate pedestrian candidate trajectories corresponding to each pedestrian risk target.

[0080] In some embodiments, Figure 2 The prediction module 204 takes the current location of the pedestrian risk target as the trajectory starting point, determines the sequence of reachable locations based on the passable area, and uses the motion boundary to constrain the state transition between adjacent reachable locations; it assigns probability weights to the state transitions based on the target confidence level to form multiple predicted trajectory branches; it performs temporal confidence screening and trajectory merging on the predicted trajectory branches to generate pedestrian candidate trajectories.

[0081] In some embodiments, Figure 2 The calculation module 205 performs time-series matching of pedestrian candidate trajectories and vehicle dynamic sweep envelopes according to the prediction time window to determine the intersection state of pedestrian risk targets entering the vehicle-occupied area; calculates the corresponding intersection time, intersection position and spatial margin according to the intersection state; and couples the intersection time, intersection position and spatial margin with the new energy braking reachability domain for evaluation to generate a collision risk characterization including braking margin and risk confidence.

[0082] In some embodiments, Figure 2 The execution module 206 determines the collision warning level corresponding to the pedestrian risk target based on the risk confidence, meeting urgency and braking margin in the collision risk characterization; it generates a coordinated warning execution command based on the collision warning level, target orientation relationship and control executable boundary; and it performs linkage control on the driver's end prompt status, the vehicle external orientation reminder status and the vehicle driving output boundary according to the coordinated warning execution command.

[0083] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0084] Figure 3 This is a schematic diagram of the electronic device 3 provided in an embodiment of this application. Figure 3 As shown, the electronic device 3 of this embodiment includes: a processor 301, a memory 302, and a computer program 303 stored in the memory 302 and executable on the processor 301. When the processor 301 executes the computer program 303, it implements the steps in the various method embodiments described above. Alternatively, when the processor 301 executes the computer program 303, it implements the functions of each module / unit in the various system embodiments described above.

[0085] Electronic device 3 can be a desktop computer, laptop, handheld computer, cloud server, or other electronic device. Electronic device 3 may include, but is not limited to, processor 301 and memory 302. Those skilled in the art will understand that... Figure 3 This is merely an example of electronic device 3 and does not constitute a limitation on electronic device 3. It may include more or fewer components than shown, or different components.

[0086] The processor 301 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0087] The memory 302 can be an internal storage unit of the electronic device 3, such as a hard disk or memory of the electronic device 3. The memory 302 can also be an external storage device of the electronic device 3, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the electronic device 3. The memory 302 can also include both internal and external storage units of the electronic device 3. The memory 302 is used to store computer programs and other programs and data required by the electronic device.

[0088] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0089] If integrated modules / units are implemented as software functional units and sold or used as independent products, they can be stored in a readable storage medium (e.g., a computer-readable storage medium). Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program may include computer program code, which may be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable storage medium may include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0090] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A real-time collision avoidance and early warning method for pedestrian collisions involving new energy freight vehicles, characterized in that, include: Collect ambient environmental perception data and vehicle operating status data of new energy trucks, and perform spatiotemporal registration on the ambient environmental perception data and the vehicle operating status data to generate near-field environmental representation and vehicle status representation. Based on the vehicle state representation, a vehicle dynamic sweep envelope that changes with the vehicle's motion posture is constructed, and a new energy braking reachability domain is generated based on the power braking state in the vehicle operating state data. Based on the near-field environment characterization, visible pedestrian targets are identified, and occlusion potential pedestrian targets are generated by combining occlusion boundaries, road traffic constraints, and target motion residues, thus obtaining a set of pedestrian risk targets; Multi-time trajectory prediction is performed on the pedestrian risk target set to generate pedestrian candidate trajectories corresponding to each pedestrian risk target; The pedestrian candidate trajectory and the vehicle dynamic sweep envelope are spatiotemporally intersected and calculated, and the collision risk characterization is determined by combining the new energy braking reachability domain. Based on the collision risk characterization, a collision avoidance warning level and a warning execution command are generated, and driving prompts, external warnings, or driving control limits are executed for the new energy truck based on the warning execution command.

2. The method according to claim 1, characterized in that, The process of collecting ambient environmental perception data and vehicle operating status data of the new energy truck, and performing spatiotemporal registration on the ambient environmental perception data and the vehicle operating status data to generate near-field environmental representations and vehicle status representations includes: The surrounding environment perception data and the vehicle operation status data are collected according to a unified time reference, and the data at different collection times are time-aligned based on vehicle motion compensation. Based on the vehicle coordinate relationship of the new energy truck, the time-aligned surrounding environment perception data is mapped to a unified vehicle coordinate system to generate near-field occupancy information; The near-field occupancy information and the vehicle operating status data are correlated with confidence levels to generate the near-field environment representation and the vehicle status representation.

3. The method according to claim 1, characterized in that, The step of constructing a vehicle dynamic sweep envelope that changes with the vehicle's motion posture based on the vehicle state representation, and generating a new energy braking reachability domain based on the power braking state in the vehicle operating state data, includes: Based on the vehicle state characterization, the attitude evolution trajectory of the new energy truck within the prediction time window is determined, and the vehicle outline constraint is temporally projected along the attitude evolution trajectory to generate the vehicle dynamic sweep envelope covering the changes in the vehicle's near-field occupancy. Based on the described dynamic braking state, energy recovery constraints, mechanical braking constraints, and load constraints affecting the vehicle's deceleration response are extracted. The deceleration boundary and parking boundary within the prediction time window are constructed, and the new energy braking reachability domain is generated.

4. The method according to claim 3, characterized in that, The step of temporally projecting the vehicle outline constraints along the attitude evolution trajectory to generate the vehicle dynamic sweep envelope covering the changes in the vehicle's near-field occupancy includes: The vehicle body outline constraints are converted into a vehicle body boundary model that changes with the vehicle's attitude. Based on the sampling time within the prediction time window, the vehicle boundary model is mapped to the pose node corresponding to the attitude evolution trajectory to generate the temporal occupied region; Based on the continuous change relationship of the vehicle body boundary between adjacent pose nodes, the temporal occupied region is expanded and the envelope is merged to obtain the vehicle dynamic sweep envelope.

5. The method according to claim 1, characterized in that, The process involves identifying visible pedestrian targets based on the near-field environment characterization, and generating occluded potential pedestrian targets by combining occlusion boundaries, road traffic constraints, and target motion residues, resulting in a pedestrian risk target set, including: The near-field environment characterization is used for explicit pedestrian identification and motion consistency verification to determine visible pedestrian targets; An occlusion potential field is constructed based on the occlusion boundary and the road traffic constraints, and the occlusion potential field is extended in a short time using the target motion residue to generate an occlusion potential pedestrian target with occurrence position constraints and motion direction constraints. The visible pedestrian targets and the occluded potential pedestrian targets are fused according to the target confidence level to obtain the pedestrian risk target set.

6. The method according to claim 1, characterized in that, The step of performing multi-time trajectory prediction on the pedestrian risk target set to generate pedestrian candidate trajectories corresponding to each pedestrian risk target includes: Based on the target state, spatial constraints, and target confidence level of each pedestrian risk target in the pedestrian risk target set, a corresponding motion intention representation is constructed. By associating the motion intention representation with the near-field environment representation, the passable area and motion boundary of each pedestrian risk target within the prediction time window are determined; Based on the passable area and the movement boundary, multi-time trajectory extrapolation is performed to generate the pedestrian candidate trajectory corresponding to each pedestrian risk target.

7. The method according to claim 6, characterized in that, The step of generating candidate pedestrian trajectories corresponding to each pedestrian risk target by performing multi-moment trajectory extrapolation based on the passable area and the movement boundary includes: Taking the current location of the pedestrian risk target as the trajectory starting point, a sequence of reachable locations is determined based on the passable area, and the state transition between adjacent reachable locations is constrained by the motion boundary. The state transition is probabilistically weighted according to the target confidence level to form multiple predicted trajectory branches; The predicted trajectory branches are filtered by time series confidence and the trajectories are merged to generate the pedestrian candidate trajectories.

8. The method according to claim 1, characterized in that, The step of performing spatiotemporal intersection calculations of the pedestrian candidate trajectory and the vehicle dynamic sweep envelope, and combining this with the new energy vehicle braking reachability domain to determine the collision risk characterization, includes: The pedestrian candidate trajectory and the vehicle dynamic sweep envelope are matched in time according to the predicted time window to determine the intersection state of the pedestrian risk target entering the vehicle-occupied area. Calculate the corresponding rendezvous time, rendezvous position, and spatial margin based on the rendezvous status; The meeting time, meeting position, and spatial margin are coupled with the new energy vehicle braking reachability domain for evaluation to generate the collision risk characterization that includes braking margin and risk confidence.

9. The method according to claim 1, characterized in that, The step of generating a collision avoidance warning level and a warning execution command based on the collision risk characterization, and executing driving prompts, external warnings, or driving control limits for the new energy truck based on the warning execution command, includes: Based on the risk confidence level, meeting urgency and braking margin in the collision risk characterization, determine the collision warning level corresponding to the pedestrian risk target; Based on the collision avoidance warning level, target orientation relationship, and control executable boundary, a collaborative warning execution command is generated; According to the aforementioned collaborative early warning execution command, the driver's terminal prompt status, the vehicle external directional reminder status, and the vehicle driving output boundary are controlled in a coordinated manner.

10. A real-time collision avoidance and early warning system for pedestrian collisions in new energy freight vehicles, characterized in that, include: The acquisition module is used to collect environmental perception data and vehicle operation status data around the new energy truck, and to perform spatiotemporal registration on the environmental perception data and the vehicle operation status data to generate near-field environmental representation and vehicle status representation. The construction module is used to construct a vehicle dynamic sweep envelope that changes with the vehicle's motion posture based on the vehicle state characterization, and to generate a new energy braking reachability domain based on the power braking state in the vehicle operating state data. The generation module is used to identify visible pedestrian targets based on the near-field environment characterization, and generate occluded potential pedestrian targets by combining occlusion boundaries, road traffic constraints and target motion residues, thereby obtaining a set of pedestrian risk targets; The prediction module is used to perform multi-time trajectory prediction on the set of pedestrian risk targets and generate pedestrian candidate trajectories corresponding to each pedestrian risk target. The calculation module is used to perform spatiotemporal intersection calculation of the pedestrian candidate trajectory and the vehicle dynamic sweep envelope, and combine the new energy braking reachability domain to determine the collision risk characterization. The execution module is used to generate a collision avoidance warning level and a warning execution command based on the collision risk characterization, and to execute driving prompts, external warnings or driving control limits for new energy trucks based on the warning execution command.

Images