Anti-collision method and device for automobile, automobile and storage medium

Abstract

The invention provides an anti-collision method and device for an automobile, the automobile and a storage medium, the method is applied to the automobile in automatic driving, and the method comprises the steps that ultra-wide-angle images around the automobile are acquired, the ultra-wide-angle images are spliced into a panoramic top view, and obstacle information in the panoramic top view is recognized; acquiring multiple groups of distance detection information according to the driving state of the automobile, and fusing the obstacle information and the distance detection information to obtain an anti-collision sensing map; the anti-collision sensing map is divided into multiple layers of safety areas according to the position of a bumper of the automobile, an anti-collision decision is made according to the driving state of the automobile and the anti-collision sensing map, and an anti-collision control instruction is obtained; and according to the anti-collision control instruction, carrying out anti-collision intervention on an automatic driving vehicle.

Description

Technical Field

[0001] This application relates to the field of autonomous driving technology, and more particularly to a collision avoidance method, device, vehicle, and storage medium for automobiles. Background Technology

[0002] With the rapid development of autonomous driving technology, vehicle collision avoidance systems have become one of the core technologies for ensuring driving safety. Existing vehicle collision avoidance technologies mainly rely on a single sensor or a simple combination of multiple sensors, such as parking assistance systems that use only ultrasonic radar or vision recognition systems that rely only on cameras. These traditional solutions have many limitations in complex environments: although ultrasonic radar is accurate in ranging, it cannot effectively identify the type of obstacle; vision systems can identify targets, but their performance degrades under poor lighting conditions; and single sensors are prone to blind spots, leading to missed detections or false alarms. Summary of the Invention

[0003] This application provides a collision avoidance method, device, vehicle, and storage medium for automobiles, for improving the collision avoidance safety of autonomous vehicles.

[0004] In a first aspect, embodiments of this application provide a collision avoidance method for automobiles, applied to automobiles in autonomous driving, the method comprising: Acquire ultra-wide-angle images of the area around the vehicle, stitch the ultra-wide-angle images into a panoramic top view, and identify obstacle information in the panoramic top view; Multiple sets of distance detection information are obtained based on the vehicle's driving status. The obstacle information and the distance detection information are fused to obtain a collision avoidance perception map. The collision avoidance perception map is divided into multiple safety zones based on the position of the vehicle's bumper. Collision avoidance decisions are made based on the vehicle's driving status and the collision avoidance perception map to obtain collision avoidance control commands. The collision avoidance control command is used to intervene in the collision avoidance of the autonomous vehicle.

[0005] Secondly, embodiments of this application provide a collision avoidance device for automobiles, applied to automobiles in autonomous driving, the collision avoidance device for automobiles comprising: The image acquisition module is used to acquire ultra-wide-angle images around the vehicle, stitch the ultra-wide-angle images into a panoramic top view, and identify obstacle information in the panoramic top view. The map generation module is used to acquire multiple sets of distance detection information based on the vehicle's driving status, and to fuse the obstacle information and the distance detection information to obtain a collision avoidance perception map. The instruction generation module is used to divide the collision avoidance perception map into multiple safety zones according to the position of the vehicle's bumper, make collision avoidance decisions based on the vehicle's driving status and the collision avoidance perception map, and obtain collision avoidance control instructions. The instruction execution module is used to intervene in the collision avoidance of the autonomous vehicle according to the collision avoidance control instruction.

[0006] Thirdly, embodiments of this application provide a vehicle for performing a collision avoidance method for a vehicle as described in any of the embodiments of this application.

[0007] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to implement the collision avoidance method for automobiles as described in any one of claims 1 to 7.

[0008] This application provides a collision avoidance method for automobiles, applied to autonomous vehicles. The method includes: acquiring ultra-wide-angle images of the vehicle's surroundings; stitching the ultra-wide-angle images into a panoramic top-down view; identifying obstacle information in the panoramic top-down view; acquiring multiple sets of distance detection information based on the vehicle's driving state; fusing the obstacle information and distance detection information to obtain a collision avoidance perception map; dividing the collision avoidance perception map into multiple safety zones based on the vehicle's bumper position; making collision avoidance decisions based on the vehicle's driving state and the collision avoidance perception map to obtain collision avoidance control commands; and intervening in collision avoidance for the autonomous vehicle based on the collision avoidance control commands. In this method, by stitching ultra-wide-angle images of the vehicle's surroundings into a panoramic top-down view, a comprehensive visual perception of the vehicle's surrounding environment is achieved. Obstacle identification on the panoramic top-down view accurately identifies the type and location information of obstacles. The generated collision avoidance perception map, generated by weighted fusion of visual obstacle information and distance detection information, utilizes a multi-layered safety zone division based on the bumper position and progressive collision avoidance decisions combined with the vehicle's driving state, making the collision avoidance control commands more precise and intelligent, and improving the accuracy of obstacle detection and collision avoidance decisions. Attached Figure Description

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

[0010] Figure 1 A schematic flowchart illustrating a collision avoidance method for automobiles provided in an embodiment of this application; Figure 2A schematic diagram of a distance management configuration table provided in an embodiment of this application; Figure 3 A schematic diagram of a sensor frame provided in an embodiment of this application; Figure 4 This is a schematic block diagram of a collision avoidance device for automobiles provided in an embodiment of this application. Detailed Implementation

[0011] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0012] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the described order. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0013] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0014] It should also be further understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0015] Please see Figure 1 , Figure 1 This is a schematic flowchart illustrating a collision avoidance method for automobiles provided in an embodiment of this application. Figure 1 The method shown is applied to vehicles in autonomous driving, and the specific steps of the collision avoidance method for vehicles include: S101-S104.

[0016] S101. Acquire ultra-wide-angle images of the area around the car, stitch the ultra-wide-angle images into a panoramic top-down view, and identify obstacle information in the panoramic top-down view.

[0017] For example, ultra-wide-angle cameras mounted around the vehicle simultaneously acquire raw image data of the surrounding environment. Due to the optical characteristics of fisheye lenses, severe radial and tangential distortions occur, necessitating geometric correction of the acquired ultra-wide-angle images. Using the intrinsic parameter matrix and distortion coefficients obtained from camera calibration, the coordinates of each pixel in the image are calculated via reverse mapping, transforming the spherical fisheye image into a standard planar image conforming to perspective projection laws. The four distortion-corrected images need to be unified to the vehicle coordinate system. A homogeneous transformation matrix is ​​used to convert the image coordinates from different cameras into world coordinates with the vehicle's geometric center as the origin. The SIFT feature point detection algorithm is used to extract stable key points in the overlapping areas of adjacent images. Correspondences between images are established through feature descriptor matching, and the RANSAC algorithm is used to remove incorrect matching points and calculate precise transformation parameters. Based on the calculated transformation relationships, the four corrected images are seamlessly stitched together. A fade-in / fade-out fusion strategy eliminates brightness differences at the stitching boundaries, generating a panoramic top-down view covering the entire environment surrounding the vehicle. The pre-built YOLOv5 object detection network receives a panoramic top-down view as input, and identifies target objects such as vehicles, pedestrians, and obstacles in the image through multi-scale feature extraction and object classification regression using a convolutional neural network. It outputs structured obstacle information containing category labels, confidence scores, and bounding box coordinates.

[0018] S102. Based on the vehicle's driving status, acquire multiple sets of distance detection information, fuse obstacle information and distance detection information to obtain a collision avoidance perception map.

[0019] For example, the vehicle's driving status is monitored in real time. Data from the vehicle speed sensor, gear position sensor, and steering angle sensor are acquired via the CAN bus. When the vehicle is moving forward at a low speed, ultrasonic sensors at the front and sides are activated for close-range detection. In reverse, all sensors operate simultaneously to achieve omnidirectional distance perception. The ultrasonic sensors emit sound wave pulses of a specific frequency and receive the echo signals reflected from the target. The target distance is calculated by measuring the time difference between transmission and reception. Simultaneously, the material properties and geometric features of the target are determined based on the intensity and spectral characteristics of the echo signal. Multiple sets of distance detection information are digitally filtered to remove environmental noise interference, and a Kalman filter algorithm is used to smooth the detection data of consecutive frames, improving the accuracy and stability of distance measurement. There are differences in detection range and accuracy between obstacle information obtained from visual recognition and distance detection information from ultrasonic detection. A spatial correlation mechanism is established to map the data from both sensors to a unified vehicle coordinate system. The weighting coefficients of the visual and distance sensors are adaptively adjusted according to the current environmental conditions. In well-lit environments, the weight of visual information is increased, while in low-light or inclement weather conditions, the reliability of the distance sensor is increased. An improved DS evidence theory is used to fuse multi-sensor data. By calculating the support and conflict of different evidence sources, a more credible comprehensive perception result is generated, forming a collision avoidance perception map that includes target location, type, and threat level.

[0020] S103. Divide the collision avoidance perception map into multiple safety zones according to the position of the car's bumper, make collision avoidance decisions based on the car's driving status and the collision avoidance perception map, and obtain collision avoidance control commands.

[0021] For example, the bumper, as the physical boundary between the vehicle and its external environment, has its geometric position information obtained through vehicle CAD data. A distance measurement coordinate system is established based on the edge positions of the front and rear bumpers. The spatial areas in the collision avoidance perception map are layered according to their distance from the bumper. The area closest to the bumper is designated as a red alert zone, representing an extremely high level of danger; medium-distance areas are designated as yellow caution zones, indicating a medium risk level; and areas further away are marked as green safety zones, indicating a low-risk state. The boundaries of these multi-layered safety zones are dynamically adjusted based on the vehicle's current speed. At high speeds, the range of each layer is expanded to provide a greater safety margin, while at low speeds or when stationary, the range is reduced to improve system sensitivity. The vehicle's driving status includes parameters such as speed, acceleration, steering angle, and braking status, which are combined with the distribution of obstacles in the collision avoidance perception map to assess the threat level. When an obstacle is located in a green safety zone, a warning mode is activated; when it enters a yellow caution zone, the warning mode is upgraded; and when an obstacle appears in a red alert zone, an emergency collision avoidance procedure is triggered. The progressive collision avoidance decision selects the appropriate control strategy based on the threat level and time urgency, gradually escalating from simple audible and visual alerts to active braking intervention, ensuring the timeliness and accuracy of the collision avoidance response, and outputting collision avoidance control commands that include control type, execution intensity, and duration.

[0022] S104. Intervene in collision avoidance for autonomous vehicles according to collision avoidance control commands.

[0023] For example, collision avoidance control commands are transmitted to the corresponding actuators via the vehicle control bus, enabling direct intervention in the operating status of the autonomous vehicle. When the command type is a warning reminder, a yellow warning sign is displayed on the in-vehicle screen, accompanied by a gentle alert sound, reminding the driver to pay attention to changes in the surrounding environment without affecting normal driving operations. Warning-level commands trigger more prominent audible and visual alarms, including flashing orange warning lights and intermittent beeping sounds, while the location and movement trend of obstacles are indicated on the display interface. Emergency collision avoidance commands activate the vehicle's active safety measures, applying a preset braking torque to the brake calipers through the electronic brake control unit, while simultaneously activating hazard lights and a continuous alarm sound to alert surrounding vehicles and pedestrians. In extremely dangerous situations, the collision avoidance control command directly takes over vehicle control, reducing power output and applying maximum braking force through the engine control unit, and activating the electronic parking brake if necessary to ensure the vehicle comes to a safe stop. During the collision avoidance intervention process, the autonomous vehicle continuously monitors the vehicle status and changes in the surrounding environment. Once the threat is eliminated, the collision avoidance measures are gradually deactivated, and normal autonomous driving mode is restored. The entire intervention process is recorded in real time, providing a basis for subsequent algorithm optimization and accident analysis.

[0024] To more clearly illustrate the technical solution of this application, the technical solution of this application will be described below through specific embodiments. It should be noted that the specific embodiments are used to expand the description of the technical solution of this application, and are not intended to limit this application.

[0025] In some embodiments, the vehicle is equipped with: a 360° panoramic host, a radar controller, multiple ultra-wide-angle cameras, and multiple distance probes, all installed around the vehicle in preset installation positions. The 360° panoramic host is electrically connected to the multiple ultra-wide-angle cameras, and the radar controller is electrically connected to the multiple distance probes. The 360° panoramic host and the radar controller are connected via a data bus.

[0026] For example, the preset installation positions are determined based on the vehicle's geometry and perception requirements. The ultra-wide-angle cameras are installed in the center of the front bumper, above the trunk lid, and below the side mirrors, respectively, to ensure that the field of view of each camera can cover a specific fan-shaped area around the vehicle and fully overlap with the field of view of the adjacent cameras.

[0027] like Figure 2 As shown, the distance sensors employ a differentiated layout strategy based on functional requirements. Multiple sensors are densely installed on the front bumper to achieve accurate distance measurement in the area ahead. The side and rear bumpers have sensor positions rationally distributed according to the vehicle's contours, ensuring blind-spot-free distance perception coverage around the entire vehicle. The installation process around the vehicle requires consideration of the body sheet metal thickness, cable routing, and waterproofing requirements. Dedicated brackets are used to fix the cameras and distance sensors in predetermined positions, while ensuring that the sensor's sensing axis maintains a standard angular relationship with the vehicle's coordinate system.

[0028] For example, such as Figure 3 As shown, the 360° panoramic host, as the core hardware for image processing, establishes an electrical connection with multiple ultra-wide-angle cameras via a high-speed data cable. It employs a parallel data transmission protocol to synchronously receive and process the raw image signals from each camera in real time. The radar controller manages the distance detection function and is electrically connected to multiple distance sensors via a dedicated communication bus. It controls the operating timing, transmission power, and receiving sensitivity of each sensor to prevent mutual interference when multiple sensors are operating simultaneously. Shielded cables and filtering circuits are used during the electrical connection process to suppress electromagnetic interference, ensuring the stability and reliability of signal transmission. Redundancy design and fault detection mechanisms further enhance the fault tolerance of the hardware connection.

[0029] The data bus (BUS) serves as a communication bridge between the two main control units, employing a high-bandwidth serial communication protocol to facilitate data exchange between the 360° panoramic host and the radar controller. Information transmitted via the data bus includes image timestamps, distance measurement results, sensor status, and fusion algorithm parameters, ensuring coordinated operation and time synchronization between the two independent processing units. A standard communication protocol and error detection mechanism are established during the data bus connection process. When a sensor malfunctions or experiences data anomalies, the faulty device can be promptly isolated while maintaining the normal operation of other sensors, ensuring the continuity and stability of the entire collision avoidance sensing function.

[0030] In some embodiments, acquiring ultra-wide-angle images of the area surrounding the vehicle body and stitching these ultra-wide-angle images into a panoramic top-down view includes: capturing images of the area around the vehicle using multiple ultra-wide-angle cameras to obtain multiple sets of ultra-wide-angle images; performing distortion restoration on the multiple sets of ultra-wide-angle images to obtain standard images; converting the spherical projection of the standard images into a planar projection and unifying them to a preset vehicle coordinate system to obtain four planar images; stitching the four planar images to generate a stitched image; adjusting the brightness and contrast of the stitched image and optimizing the stitching boundaries of the stitched image to obtain a panoramic top-down view.

[0031] For example, multiple ultra-wide-angle cameras are synchronously activated according to a preset time sequence to capture images of the environment surrounding the vehicle. The image sensor converts the optical signals into digital image data and automatically adjusts exposure parameters to adapt to different lighting conditions, thus obtaining ultra-wide-angle images. Due to radial and tangential distortion caused by the lens's optical characteristics, reverse distortion correction calculations are performed on each pixel in multiple sets of ultra-wide-angle images using a pre-calibrated camera intrinsic parameter matrix and distortion coefficients. Pixel holes during the correction process are filled using bilinear interpolation while maintaining image edge sharpness, resulting in a standard image with distortion eliminated. The spherical projection of the standard image is converted to a planar projection through a mathematical transformation from a spherical coordinate system to a Cartesian coordinate system. Using the fixed installation relationship between each camera and the vehicle body and a homogeneous transformation matrix, the images from different camera coordinate systems are uniformly transformed to a coordinate system with the vehicle's geometric center as the origin, forming four planar images with the same coordinate reference. A scale-invariant feature transform algorithm is used to extract stable key points in the overlapping areas of adjacent images and establish feature correspondences. Weighted averaging and gradient fusion techniques are used to handle brightness differences and geometric deviations at the stitching boundaries. Four planar images are merged according to their spatial relationships to generate a stitched image covering the complete environment surrounding the vehicle. Histogram equalization and adaptive contrast enhancement algorithms are used to adjust the overall brightness and local contrast of the stitched image. Multi-resolution fusion and boundary smoothing algorithms are employed to further eliminate stitching artifacts and ensure natural transitions in boundary areas, resulting in a panoramic top-down view.

[0032] In some embodiments, acquiring multiple sets of distance detection information based on the vehicle's driving state includes: activating a preset group of distance probes when the vehicle is in a forward driving state and its speed is within a preset low-speed range; activating all distance probes when the vehicle is in a reverse driving state; transmitting distance detection signals and receiving reflected echoes through the distance probes in the working state, and calculating the target distance based on the reflected echoes; preprocessing the reflected echoes to obtain the signal strength of the reflected echoes, and determining the object's material and size characteristics based on the signal strength; mapping the target distance, signal strength, object material, and size characteristics to the corresponding spatial grid in a three-dimensional raster map to obtain multiple sets of distance detection information.

[0033] For example, the vehicle's driving status is determined in real time by reading data from the vehicle speed sensor, gear shift switch, and steering wheel angle sensor on the CAN bus. When the vehicle is detected to be moving forward and its speed is within a preset low-speed range, preset groups of distance sensors in the central area of ​​the front bumper and on both sides are activated according to the detection requirements of the vehicle's direction of travel. The selection of preset groups is based on the main threat direction when the vehicle is moving forward. The front and side-front sensor combinations are selectively activated through control signals, while the rear sensors are temporarily turned off to reduce power consumption and unnecessary data processing burden. When the vehicle switches to reverse mode, the gear shift sensor detects the reverse gear signal and immediately sends a working command to all distance sensors through the radar controller, activating all sensors around the vehicle to obtain complete environmental perception capabilities.

[0034] In operation, the distance probes sequentially emit distance detection signals of specific frequencies according to a preset timing control scheme, avoiding signal crosstalk and ranging errors that occur when multiple probes emit simultaneously. Each probe's emitted ultrasonic pulse generates a reflected echo upon encountering an obstacle. The probe's built-in receiving transducer captures the returned reflected signal and converts it into an electrical signal, which is then transmitted to the radar controller. The propagation time of the reflected echo is measured using a high-precision timer, and the accurate distance between the probe and the target obstacle is calculated using the known speed of sound. Multiple measurements are then processed using a moving average filtering algorithm to improve the stability and reliability of the target distance measurement.

[0035] The reflected echo signal carries rich target feature information during transmission. The amplitude, spectrum, and waveform characteristics of the echo are extracted through analog-to-digital conversion and digital signal preprocessing. Signal strength analysis is based on the variation pattern of the echo amplitude. Different target surfaces exhibit significantly different ultrasonic wave reflection characteristics; metallic surfaces produce strong specular reflection, while fabrics or foams show strong absorption characteristics. Based on the signal strength attenuation and spectral distribution characteristics, combined with a pre-defined material identification algorithm, the target's material properties are determined. Simultaneously, the target's geometric dimensions and shape characteristics are inferred through echo pulse width and multipath reflection analysis.

[0036] The target distance, azimuth, and material feature information obtained by each probe need to be uniformly transformed into the vehicle coordinate system. Through coordinate transformation, the detection results in polar coordinate form are mapped to a rectangular coordinate system with the vehicle's geometric center as the origin. The 3D grid map divides the space around the vehicle into regular grid cells. Each spatial grid corresponds to a specific coordinate range and height level. The probe's detection results are assigned to the corresponding grid and marked with the occupancy status according to the target location information.

[0037] In some embodiments, obstacle information and distance detection information are fused to obtain a collision avoidance perception map, including: determining weight allocation parameters based on ambient light intensity and weather conditions; assigning confidence levels to obstacle information and distance detection information based on the weight allocation parameters to obtain multi-sensor data; performing conflict detection and consistency verification on the multi-sensor data to obtain conflict identification data; optimizing the overlapping detection area based on the conflict identification data to obtain perception data, wherein the overlapping detection area is the area where the detection range of the ultra-wide-angle camera and the detection range of multiple distance probes overlap; and performing spatial mapping and region division on the perception data to obtain a collision avoidance perception map.

[0038] For example, the ambient light intensity is measured in real time by an onboard light sensor, and combined with a weather recognition algorithm to determine the impact of different weather conditions, such as sunny, cloudy, rainy, or foggy days, on the performance of the visual sensor. The weighting parameters are determined based on the principle of environmental adaptability. Under ideal conditions with sufficient lighting, the weight of obstacle information acquired by the visual sensor is increased, while under low light, backlight, or adverse weather conditions, the weight of the reliability of distance sensor data is increased. The weighting parameters are dynamically adjusted using a lookup table method or interpolation calculation to avoid a decrease in fusion performance caused by fixed weights when the environment changes.

[0039] DS evidence theory, as a mathematical framework for multi-sensor data fusion, treats the detection results of different sensors as independent sources of evidence, assigning a corresponding confidence level to each piece of evidence through a basic probability assignment function. Obstacle information originates from visual recognition results, including target category, location, and confidence level information, while distance detection information provides accurate distance measurement and material feature judgment. Based on pre-defined weighting parameters, different basic probability qualities are assigned to visual and radar evidence, forming multi-sensor data. The multi-sensor data structure contains multiple focal elements and corresponding confidence levels.

[0040] In the process of multi-sensor data fusion, information conflicts between different evidence sources are inevitable. The degree of inconsistency is quantified by calculating the conflict coefficient between the evidence. Conflict degree detection uses the K-value calculation method to assess the severity of evidence conflict. When the conflict degree exceeds a preset threshold, it indicates that there is a significant discrepancy in the sensor data and further processing is required. Consistency verification identifies anomalous data and sensor malfunctions through cross-validation and time correlation analysis, and the detection results are marked in the data structure to form fused data containing conflict markers.

[0041] The overlapping detection region is the intersection of the ultra-wide-angle camera's field of view and the fan-shaped detection region of the range probe. Within this region, both sensors can acquire target information, but their accuracy and feature description differ. Based on the anomaly markers in the conflict identification data, the multi-sensor detection results within the overlapping region are weighted and filtered, prioritizing sensor data with high confidence and good consistency with other evidence. A Bayesian update mechanism is then used to correct uncertainties and eliminate contradictory data.

[0042] Spatial mapping of the perceived data transforms target information from different sensor coordinate systems to the vehicle's body coordinate system, handling coordinate rotation and translation relationships through a homogeneous transformation matrix. Region division categorizes and labels targets based on their spatial location, motion state, and threat level, discretizing continuous spatial locations into clearly defined regional units. Each region contains comprehensive information such as target type, distance, relative speed, and threat rating, forming a collision avoidance perception map.

[0043] Spatial mapping of the perceived data transforms target information from different sensor coordinate systems to the vehicle's body coordinate system, handling coordinate rotation and translation relationships through a homogeneous transformation matrix. Region division categorizes and labels targets based on their spatial location, motion state, and threat level, discretizing continuous spatial locations into clearly defined regional units. Each region contains comprehensive information such as target type, distance, relative speed, and threat rating, forming a collision avoidance perception map.

[0044] In a specific embodiment, spatial mapping and regional division of the sensing data are performed to obtain a collision avoidance sensing map, including: matching and associating obstacle information with distance detection information within overlapping detection areas by calculating the intersection-union ratio (IU) of bounding boxes; identifying new obstacle targets when the IU is lower than a preset threshold to obtain first sensing data; grouping the first sensing data into regions according to the spatial distribution of distance detectors; verifying the data in each detection area by selecting the detection result with the highest confidence based on signal strength and target distance to obtain second sensing data; mapping the obstacle position and distance information in the second sensing data to the spatial grid of a three-dimensional raster map; transforming the data from different detector coordinate systems to the vehicle coordinate system through coordinate transformation to obtain third sensing data; performing spatial clustering analysis on the third sensing data; merging similar obstacle information in adjacent spatial grids; and identifying obstacle types based on object material and size characteristics to obtain fourth sensing data; and dividing the three-dimensional raster map into different protection zones according to the degree of safety risk based on the threat level and spatial distribution characteristics of obstacles in the fourth sensing data to obtain a collision avoidance sensing map.

[0045] The intersection-over-union ratio (IoU) of the visual bounding boxes of obstacles and the spatial bounding boxes of distance detection information within the overlapping detection area is calculated. The consistency of the two sensor data in identifying the same target is evaluated through spatial geometric overlap measurement. When the IoU exceeds a set threshold, visual and distance features are correlated; otherwise, an independent tracking marker is established for new obstacle targets. The IoU calculation considers the three-dimensional spatial occupancy of obstacles and differences in sensor detection accuracy, dynamically adjusting the matching threshold to adapt to target recognition needs under various environmental conditions. This forms the first perception data, including correlated markers and new target labels, effectively solving the problems of missed detections and duplicate identifications by a single sensor. The first perception data is spatially grouped according to the physical installation location of the distance probe around the vehicle and the detection sector coverage area. Each detection area corresponds to an environmental perception responsibility area within a specific azimuth angle range. Within each detection area, a confidence assessment mechanism is established by comprehensively analyzing the attenuation characteristics of reflected echo signal intensity and target distance values. The detection results with the highest confidence were obtained through multiple verification criteria, including signal-to-noise ratio analysis, multi-frame data consistency verification, and cross-validation of data from adjacent probes. This ensured the stability and accuracy of the selected data. The data verification process considered the impact of probe aging, environmental interference, and target surface material on signal reflection characteristics. Adaptive filtering and outlier removal algorithms were used to improve data quality, resulting in second-sensor data that significantly improved the accuracy and reliability of distance measurement. The verified obstacle spatial coordinates and accurate distance measurement results from the second-sensor data were allocated to corresponding spatial grid cells in the 3D raster map according to a predefined grid resolution. A homogeneous coordinate transformation matrix was used to uniformly transform the detection data from the local coordinate systems of different distance probes to the standard vehicle coordinate system with the vehicle's geometric center as the origin. The coordinate transformation process involved geometric operations such as rotation, translation, and scaling to ensure that data from different probes maintained a consistent spatial reference. The 3D raster map adopted a regular cubic structure, with each grid cell storing attribute information such as the occupancy status, obstacle type, and confidence level of that spatial location, forming structured third-sensor data and achieving a unified spatial representation of multi-sensor data. Density-based spatial clustering is performed on the third-sensor data to identify sets of spatially adjacent and similarly attributed grid cells in the 3D raster map. These cells are then classified as the same obstacle entity. The spatial clustering process controls the density and size of clusters by setting neighborhood radii and minimum point thresholds. Obstacle information within adjacent spatial grids is merged, taking into account the confidence weights of each grid. A weighted average method is used to calculate the representative location and size parameters of the merged obstacles. Based on the object material judgment results from distance detection information and the size feature information obtained from multiple grids, a corresponding type label is assigned to each clustered obstacle, forming fourth-sensor data with clear target classification, thus avoiding fragmented target identification and duplicate counting problems.Threat levels are assessed based on the classification, motion state, and spatial relationship of obstacles in the fourth-sensor data. A threat level classification system from low to high risk is established, with the criteria for judging the threat level including a comprehensive assessment of multiple dimensions such as the relative distance between the obstacle and the vehicle, relative speed, collision probability, and potential damage. Spatial distribution characteristics are analyzed to statistically determine the number density, type distribution, and motion trends of obstacles in different areas, identifying high-risk areas requiring special attention and relatively safe passage areas. The 3D grid map area division process divides the entire sensing space into different protection zones according to the degree of safety risk, such as emergency braking zones, warning zones, and normal monitoring zones, based on the threat level assessment results. Each zone corresponds to a specific collision avoidance response strategy and control measures, forming a collision avoidance sensing map that supports layered collision avoidance decision-making. This achieves a complete transformation process from raw sensing data to structured collision avoidance decision support information.

[0046] In some embodiments, the collision avoidance perception map is divided into multiple safety zones based on the position of the vehicle's bumper, including: calibrating distance reference points on the collision avoidance perception map based on the position of the vehicle's bumper to obtain a distance reference map; marking the areas near the bumper position in the distance reference map with red warning signs and 3D transparent red lines to obtain a first zone map; marking the areas at a medium distance from the bumper position in the first zone map with yellow caution signs and 3D transparent yellow lines to obtain a second zone map; marking the areas farther from the bumper position in the second zone map with green safety signs and 3D transparent green lines to obtain a third zone map; acquiring the vehicle's steering wheel angle data, and dynamically overlaying and updating the third zone map in real time based on the steering wheel angle data to obtain multiple safety zones.

[0047] For example, the precise geometric position of a car's bumper in the vehicle coordinate system is determined through vehicle CAD data and actual measurements, including the boundary coordinates and height information of the front and rear bumpers. The distance reference point calibration process uses the bumper edge as the starting reference point for distance measurement, establishing a distance coordinate system with the bumper position as the origin in the collision avoidance perception map's data structure. The calibration process considers the influence of vehicle geometric changes and loading conditions, ensuring the accuracy of the distance reference through dynamic calibration, resulting in a reference map data structure with a clear distance reference.

[0048] The delineation of adjacent zones is based on vehicle braking distance and collision risk assessment, defining the spatial range closest to the bumper as an extremely high-risk danger zone. All spatial grids within this zone are marked with red alerts, highlighting the threat level of high-risk areas through color coding and priority marking. Three-dimensional transparent red line markers are used to draw red zone boundaries on the collision avoidance map using 3D visualization technology. The lines have a certain degree of transparency to avoid obscuring other important information while maintaining sufficient display intensity to ensure the driver can clearly identify the danger zone. The first-zone map overlays the red alert zone marker information onto the original collision avoidance map, forming an updated data structure containing clearly defined boundaries of high-risk areas.

[0049] The medium-distance zone is designed with vehicle reaction time and braking performance parameters in mind, defining an appropriate area outside the red alert zone as a warning zone requiring attention. The yellow warning indicator uses a different color code and display priority than the red zone, with a gradient color indicating a decreasing risk level. A three-dimensional transparent yellow line marks the yellow zone boundary outside the red border of the first zone map, maintaining visual harmony and distinction with the red line in terms of line style and transparency. The second zone map integrates the dual identification information of the red alert zone and the yellow warning zone, establishing a hierarchical risk level display structure.

[0050] The more distant area encompasses the detection range beyond the yellow warning zone, representing a relatively safe area that still requires monitoring. The green safety marker uses green color coding, representing safety and passage, and its priority is set to the lowest to avoid interfering with attention to high-risk areas. A three-dimensional transparent green line marks the outermost boundary of the area, forming a complete three-layered concentric area structure with the inner red and yellow lines. The third area map integrates the complete safety zone division results of the three color markers, establishing a progressive risk level display system from high-risk to safe.

[0051] Steering wheel angle data is acquired in real time via an angle sensor, reflecting the driver's steering intention and the vehicle's expected trajectory. Dynamic trajectory overlay calculates the future driving path based on the current angle value and vehicle kinematics, overlaying the predicted trajectory as a dynamic curve onto the third-zone map display. A real-time update mechanism continuously monitors changes in angle data and adjusts the shape and position of the trajectory lines accordingly, ensuring the displayed driving path always reflects the current driving status. Multiple safety zones, under the influence of the dynamic trajectory lines, form a comprehensive display effect with both directionality and predictive capabilities, providing the driver with intuitive route planning and risk assessment information.

[0052] In some embodiments, the vehicle further includes: a warning and alert unit, an active warning unit, an emergency braking unit, and an automatic collision avoidance unit. These units make collision avoidance decisions based on the vehicle's driving status and a collision avoidance perception map, obtaining collision avoidance control commands, including: locating obstacles in the collision avoidance perception map; activating the warning and alert unit to display a yellow trajectory line when the obstacle enters a green area within a multi-layered safety zone, obtaining a Level 1 warning signal; activating the active warning unit to display an orange indicator when the obstacle enters a yellow attention area within the multi-layered safety zone, obtaining a Level 2 warning signal; activating the emergency braking unit to preload the braking system when the obstacle enters a red alert area within the multi-layered safety zone, obtaining a Level 3 emergency control signal; triggering the automatic collision avoidance unit to perform emergency braking based on the Level 3 emergency control signal and the rate of change of the obstacle's location, obtaining a Level 4 automatic collision avoidance signal; and optimizing the threshold and adjusting the sensitivity of the Level 4 automatic collision avoidance signal based on the driver's historical operation data and current environmental characteristic parameters, obtaining the collision avoidance control commands.

[0053] For example, the location of obstacles in the collision avoidance perception map is determined by the output of an object detection algorithm. Each identified obstacle has specific spatial coordinates, geometric dimensions, and motion state information. The calibration process compares this location information with the boundary coordinates of multiple safety zones to determine the current risk level zone of the obstacle and its precise distance relative to the vehicle bumper.

[0054] When an obstacle enters the green safety zone, the threat level is assessed as low-risk, but monitoring and early warning are still required. Upon receiving the activation command, the warning unit initiates the yellow trajectory line display function, drawing the expected yellow driving trajectory on the in-vehicle display screen and marking the relative position of the obstacle. The trajectory line display uses a smooth curve, and the color saturation and line thickness are optimized to ensure visibility under different lighting conditions. The Level 1 warning signal, as the most basic level of alert, informs the driver of changes in the surrounding environment through visual cues, maintaining necessary safety awareness while avoiding excessive intervention in normal driving operations.

[0055] When an obstacle advances into the yellow warning zone, the risk level escalates to moderate threat, requiring more prominent warning measures. Upon receiving the escalation activation command, the active warning unit immediately initiates the orange indicator display program, displaying an orange warning icon and obstacle location indicator in a prominent position on the display screen. The orange indicator uses flashing or animated effects to enhance visual impact, while a slight vibration alert or audio alarm ensures the driver can promptly perceive the change in threat level. The intensity and duration of the secondary warning signal are dynamically adjusted based on the obstacle's approach speed and dwell time, avoiding excessive sensory stimulation while ensuring effective warning.

[0056] The presence of an obstacle within the red alert zone indicates an extremely high risk of collision, requiring immediate activation of emergency protective measures. Upon receiving the activation command, the emergency braking unit begins preloading the braking system, shortening the response delay time by pressurizing the hydraulic pump and pre-fitting the brake pads. During the preloading process, sufficient pre-pressure is maintained without producing a noticeable braking effect, avoiding sudden deceleration that could cause discomfort to the driver and passengers. The Level 3 emergency control signal simultaneously activates all available warning and alert devices, including multimodal alerts such as high-brightness flashing lights, continuous siren sounds, and seat vibrations, ensuring the driver is aware of the emergency in the shortest possible time.

[0057] The rate of change of obstacle position is calculated through position difference between consecutive frames, reflecting the relative speed and acceleration of the target approaching the vehicle. The Level 3 emergency control signal, combined with the obstacle's motion trend, predicts the collision time. When the predicted collision time is less than the safety threshold and the relative speed exceeds the danger limit, the automatic collision avoidance unit is triggered. The automatic collision avoidance unit performs emergency braking, directly applying the maximum safe braking force through the electronic brake controller while simultaneously reducing engine output power. The Level 4 automatic collision avoidance signal represents the highest level of response from the vehicle's autonomous safety functions. While performing emergency braking, it records sensor data and vehicle status information before and after the event for subsequent analysis.

[0058] Driver historical operation data is obtained through long-term recording and analysis, including personalized characteristics such as braking habits, steering characteristics, and reaction patterns to warning signals. Current environmental characteristics parameters cover external factors affecting driving safety, such as road conditions, weather, traffic density, and lighting conditions. The threshold optimization process uses machine learning algorithms to analyze the correlation between historical data and environmental parameters, dynamically adjusting the triggering conditions and response intensity of alarms at all levels. Sensitivity adjustment is personalized based on the driver's driving style and environmental adaptability, reducing false alarms and excessive intervention while ensuring safety. Collision avoidance control commands integrate multi-level threat assessment results and personalized adjustment parameters, providing precise control guidance to vehicle actuators.

[0059] Please see Figure 4 , Figure 4 This is a schematic block diagram of a collision avoidance device for automobiles provided in an embodiment of this application. The collision avoidance device 200 for automobiles is used to perform the aforementioned collision avoidance method for automobiles and is applied to automobiles in autonomous driving. The collision avoidance device 200 for automobiles can be configured in a server.

[0060] The server can be a standalone server, a server cluster, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0061] like Figure 4 As shown, the collision avoidance device 200 for automobiles includes: an image acquisition module 201, a map generation module 202, an instruction generation module 203, and an instruction execution module 204.

[0062] The image acquisition module 201 is used to acquire ultra-wide-angle images of the area around the car, stitch the ultra-wide-angle images into a panoramic top view, and identify obstacle information in the panoramic top view.

[0063] The map generation module 202 is used to acquire multiple sets of distance detection information based on the vehicle's driving status, and to fuse obstacle information and distance detection information to obtain a collision avoidance perception map.

[0064] The instruction generation module 203 is used to divide the collision avoidance perception map into multiple safety zones according to the position of the car's bumper, make collision avoidance decisions based on the car's driving status and the collision avoidance perception map, and obtain collision avoidance control instructions.

[0065] The instruction execution module 204 is used to intervene in collision avoidance for autonomous vehicles according to collision avoidance control instructions.

[0066] This application provides a vehicle for performing a collision avoidance method for a vehicle as described in any of the embodiments of this application.

[0067] This application provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program causes the processor to implement a collision avoidance method for automobiles as described in any of the embodiments of this application.

[0068] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A collision avoidance method for automobiles, characterized in that, The method, applied to vehicles used in autonomous driving, includes: Acquire ultra-wide-angle images of the area around the vehicle, stitch the ultra-wide-angle images into a panoramic top view, and identify obstacle information in the panoramic top view; Multiple sets of distance detection information are obtained based on the vehicle's driving status. The obstacle information and the distance detection information are fused to obtain a collision avoidance perception map. The collision avoidance perception map is divided into multiple safety zones based on the position of the vehicle's bumper. Collision avoidance decisions are made based on the vehicle's driving status and the collision avoidance perception map to obtain collision avoidance control commands. The collision avoidance control command is used to intervene in the collision avoidance of the autonomous vehicle.

2. The collision avoidance method for automobiles as described in claim 1, characterized in that, The vehicle is equipped with: a 360-degree panoramic host, a radar controller, multiple ultra-wide-angle cameras, and multiple distance probes, all installed around the vehicle in preset installation positions. The 360-degree panoramic host is electrically connected to the multiple ultra-wide-angle cameras, and the radar controller is electrically connected to the multiple distance probes. The 360-degree panoramic host and the radar controller are connected via a data bus.

3. The collision avoidance method for automobiles as described in claim 2, characterized in that, The step of acquiring an ultra-wide-angle image of the area surrounding the vehicle and stitching the ultra-wide-angle image into a panoramic top-down view includes: Multiple ultra-wide-angle cameras are used to capture images of the area around the vehicle, resulting in multiple sets of ultra-wide-angle images. Distortion restoration is performed on multiple sets of the ultra-wide-angle images to obtain standard images; The spherical projection of the standard image is converted into a planar projection and unified to a preset vehicle coordinate system to obtain four planar images; The four planar images are stitched together to generate a stitched image; Adjust the brightness and contrast of the stitched image and optimize the stitching boundary to obtain a panoramic top view.

4. The collision avoidance method for automobiles as described in claim 2, characterized in that, The process of acquiring multiple sets of distance detection information based on the vehicle's driving status includes: When the vehicle is in a forward driving state and its speed is within a preset low speed range, the distance probe of the preset group is activated. When the vehicle is in reverse, all distance sensors are activated. The distance probe, in its working state, transmits a distance detection signal and receives the reflected echo, and calculates the target distance based on the reflected echo; The reflected echo is preprocessed to obtain the signal intensity of the reflected echo, and the material and size characteristics of the object are determined based on the signal intensity. The target distance, signal strength, object material, and size features are mapped to the corresponding spatial grid in the three-dimensional raster map to obtain multiple sets of distance detection information.

5. The collision avoidance method for automobiles as described in claim 2, characterized in that, The process of fusing the obstacle information and the distance detection information to obtain a collision avoidance perception map includes: The weighting parameters are determined based on ambient light intensity and weather conditions. The obstacle information and the distance detection information are assigned confidence levels according to the weighting parameters to obtain multi-sensor data. The conflict degree of the multi-sensor data is detected and the consistency is verified to obtain conflict identification data; Based on the conflict identification data, the overlapping detection area is optimized to obtain perception data. The overlapping detection area is the overlapping area of ​​the detection range of the ultra-wide-angle camera and the detection range of multiple distance probes. Spatial mapping and regional division are performed on the perceived data to obtain a collision avoidance perception map.

6. The collision avoidance method for automobiles as described in claim 2, characterized in that, The step of dividing the collision avoidance perception map into multiple safety zones based on the position of the vehicle's bumper includes: The distance reference point on the collision avoidance perception map is marked according to the position of the car's bumper to obtain the distance reference map; In the distance reference map, the area adjacent to the location of the bumper is marked with a red warning sign and a three-dimensional transparent red line to obtain a first area map; A second area map is obtained by marking the area at a medium distance from the bumper in the first area map with yellow warning signs and three-dimensional transparent yellow lines; In the second area map, areas farther away from the bumper are marked with green safety markers and three-dimensional transparent green lines to obtain a third area map; The steering wheel angle data of the car is obtained, and the third area map is dynamically overlaid and updated in real time based on the steering wheel angle data to obtain a multi-layered safety zone.

7. The collision avoidance method for automobiles as described in claim 6, characterized in that, The vehicle also includes: a warning and alert unit, an active warning unit, an emergency braking unit, and an automatic collision avoidance unit. The collision avoidance decision is made based on the vehicle's driving status and the collision avoidance perception map, resulting in collision avoidance control commands, including: The locations of obstacles in the collision avoidance perception map are identified; When the obstacle enters the green area of ​​the multi-layered safety zone, the warning unit is activated to display a yellow trajectory line, thus obtaining a first-level warning signal. When the obstacle enters the yellow warning area of ​​the multi-layered safety zone, the active warning unit is upgraded and activated to display an orange indicator, resulting in a level two warning signal. When the obstacle enters the red alarm zone of the multi-layered safety area, the emergency braking unit is activated to preload the braking system and obtain a level three emergency control signal. Based on the level three emergency control signal and the rate of change of the obstacle's position, the automatic collision avoidance unit is triggered to perform emergency braking, resulting in a level four automatic collision avoidance signal. Based on the driver's historical operation data and current environmental characteristic parameters, the threshold of the Level 4 automatic collision avoidance signal is optimized and the sensitivity is adjusted to obtain the collision avoidance control command.

8. A collision avoidance device for automobiles, characterized in that, For use in autonomous vehicles, the collision avoidance device for vehicles is used to perform the collision avoidance method for vehicles as described in any one of claims 1-7, and for use in autonomous vehicles, the collision avoidance device for vehicles includes: The image acquisition module is used to acquire ultra-wide-angle images around the vehicle, stitch the ultra-wide-angle images into a panoramic top view, and identify obstacle information in the panoramic top view. The map generation module is used to acquire multiple sets of distance detection information based on the vehicle's driving status, and to fuse the obstacle information and the distance detection information to obtain a collision avoidance perception map. The instruction generation module is used to divide the collision avoidance perception map into multiple safety zones according to the position of the vehicle's bumper, make collision avoidance decisions based on the vehicle's driving status and the collision avoidance perception map, and obtain collision avoidance control instructions. The instruction execution module is used to intervene in the collision avoidance of the autonomous vehicle according to the collision avoidance control instruction.

9. A car, characterized in that, Used to perform the collision avoidance method for automobiles as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to implement the collision avoidance method for automobiles as described in any one of claims 1 to 7.

Images