Automobile obstacle early warning system based on edge calculation

Abstract

The invention relates to an automobile obstacle early warning system based on edge calculation. The system comprises a data acquisition module, an edge calculation module and an early warning module which are sequentially connected through a data transmission bus, the data acquisition module is used for acquiring the multi-source data of the surrounding environment of the vehicle acquired by the sensor module; the multi-source data comprises obstacle data, multispectral image data and environment data; the edge calculation module is used for performing fusion calculation on the multi-source data and performing obstacle recognition according to an obstacle recognition model to obtain an obstacle type and a corresponding obstacle danger degree; the early warning module is used for generating a corresponding early warning signal and an early warning execution instruction according to the danger degree of the obstacle; the early warning execution instruction is used for indicating the automobile early warning component to work according to the early warning signal. By adopting the system, the overall early warning response speed can be greatly increased, and the obstacle detection precision in a complex road environment can be greatly improved.

Description

Technical Field

[0001] The present invention belongs to the field of automobile driving safety assistance, and in particular relates to an automobile obstacle warning system based on edge computing. Background Art

[0002] With the development of automobile technology, intelligent assisted driving technology has emerged. By using various sensors, advanced algorithms and intelligent control units, it can achieve real-time monitoring and analysis of the vehicle's surrounding environment, thereby providing the driver with auxiliary decision-making information, significantly improving driving safety and convenience.

[0003] Traditional obstacle detection and warning systems rely primarily on millimeter-wave radar and cameras. Millimeter-wave radar transmits and receives millimeter-wave signals to calculate the distance, speed, and angle of the target object; cameras collect visual images of the vehicle's surroundings. A data processing unit receives and analyzes the combined data. If a hazard is detected, the warning execution unit issues an audible and visual alarm.

[0004] However, the obstacle detection and warning systems used by millimeter-wave radars and cameras are significantly affected by complex environments like inclement weather, significantly reducing detection range and accuracy and making them prone to misjudgments or missed detections. Regarding the timeliness of warnings, existing technologies have a complex data processing process. There is a certain delay between sensor data acquisition and the processing unit analyzing and issuing warnings. At high speeds, the driver's reaction time is extremely short, making it impossible to issue timely warnings. Current systems are mostly designed for conventional road scenarios, and their performance degrades significantly in special scenarios such as unpaved roads and extreme weather conditions. Summary of the Invention

[0005] Based on this, it is necessary to provide an edge computing-based automobile obstacle warning system that can improve detection accuracy and response speed to address the above technical problems.

[0006] In a first aspect, the present application provides an obstacle warning system for automobiles based on edge computing: the system includes a data acquisition module, an edge computing module, and a warning module connected in sequence via a data transmission bus;

[0007] The data acquisition module is used to obtain multi-source data of the vehicle's surrounding environment collected by the sensor module; the multi-source data includes obstacle data, multispectral image data and environmental data;

[0008] The edge computing module is used to perform fusion calculations on multi-source data and identify obstacles based on the obstacle recognition model to obtain the obstacle type and the corresponding obstacle danger level;

[0009] The warning module is used to generate corresponding warning signals and warning execution instructions according to the degree of danger of obstacles; the warning execution instructions are used to instruct the automobile warning components to work according to the warning signals.

[0010] In one embodiment, the data acquisition module includes a millimeter wave radar sensing unit, a multispectral camera driving unit, and a sensor module unit;

[0011] The millimeter-wave radar sensing unit is used to calculate the time difference and frequency change between the transmitted wave and the reflected wave by frequency modulating the continuous wave to obtain obstacle data; the obstacle data includes distance data, speed data and angle data;

[0012] The multispectral camera driving unit is used to drive the multispectral camera to collect visible light images and infrared light images of the vehicle's surrounding environment at a preset frame rate, and preprocess the obtained visible light images and infrared light images to obtain multispectral image data; the preprocessing corresponding steps include denoising and enhancement processing;

[0013] The sensor module unit is used to perform standardized numerical conversion processing on the analog signals of external light intensity value, temperature value and humidity collected by the sensor module to obtain environmental data.

[0014] In one embodiment, the edge computing module includes a data fusion unit and an obstacle identification unit;

[0015] The data fusion unit is used to fuse obstacle data and multispectral data through a data fusion algorithm and optimize it with environmental data to obtain obstacle environment fusion data;

[0016] The obstacle recognition unit is used to input the obstacle environment fusion data into the pre-trained obstacle recognition model to obtain the obstacle type, and determine the obstacle danger level based on the vehicle's driving status, combined with the obstacle type and the obstacle's position and speed data; the obstacle's position is determined by distance data and angle data.

[0017] In one embodiment, obstacle data and multispectral data are fused by a data fusion algorithm and optimized using environmental data to obtain obstacle environment fusion data, including:

[0018] The distance data, speed data and angle data of the obstacle data are processed through multiple convolutional layer features to obtain the time series speed-distance graph features;

[0019] The structure contour of multispectral image data is extracted by ResNet50 and asymmetric attention module to obtain contour features;

[0020] The cross-modal attention mechanism is used to unify the temporal speed-distance graph features and contour features, and then optimized in combination with the environmental data to obtain obstacle environment fusion data.

[0021] In one embodiment, the obstacle recognition model obtains the obstacle type by the following method:

[0022] Extract the entity form, dynamic state and risk situation of obstacles based on the fusion data of obstacle environment;

[0023] Classify the category labels according to the entity shape, dynamic state and risk situation to obtain the obstacle type.

[0024] In one embodiment, the obstacle danger level is obtained by:

[0025] Based on the driving status of the car, the obstacle type is scored to obtain a danger score;

[0026] The hazard scores are classified into levels according to the preset threshold interval to obtain the obstacle hazard level; the obstacle hazard level includes low risk, medium risk and high risk;

[0027] The risk score is obtained using the following formula:

[0028] Risk Score=α·T r ·V r ·P r ·E r

[0029]

[0030] Among them, RiskScore is the risk score; T r is the obstacle type risk factor; V r is a dynamic predictor; P r Position information factor; E r is the environmental disturbance factor; α is the adjustment factor; k is the speed control growth rate; V rel is the relative speed between the obstacle and the car; v th is the dangerous speed threshold.

[0031] In one embodiment, the early warning module includes an early warning signal generating unit and an early warning executing unit;

[0032] The warning signal generation unit is used to generate corresponding warning signals according to the preset warning strategy based on the danger level of the obstacle; the warning signals include light flashing signals, sound signals and seat vibration signals;

[0033] The warning execution unit is used to generate multi-dimensional warning execution instructions based on the warning signal; the warning execution instructions include a light instruction for instructing an indicator light of a control panel to flash according to a light flashing signal, a sound instruction for instructing a power amplifier circuit to drive a buzzer according to a sound signal, and a vibration instruction for instructing a pulse width modulation to drive a vibration electrode to vibrate according to a seat vibration signal;

[0034] Among them, the preset early warning strategy corresponds to the following steps:

[0035] When the obstacle hazard level is medium risk, the warning signals include a light flashing signal at a preset low frequency threshold, a sound signal at a preset medium threshold intensity, and a seat vibration signal at a preset light threshold intensity;

[0036] When the obstacle danger level is high risk, the warning signal includes a light flashing signal of a preset high frequency threshold, a sound signal of a preset high threshold intensity, and a seat vibration signal of a preset strong threshold intensity.

[0037] Secondly, this application also provides an automobile obstacle warning method based on edge computing, including:

[0038] Acquire multi-source data of the vehicle's surrounding environment collected by the sensor module; the multi-source data includes obstacle data, multispectral image data, and environmental data;

[0039] Perform fusion calculation on multi-source data and identify obstacles based on the obstacle recognition model to obtain the obstacle type and the corresponding obstacle danger level;

[0040] The corresponding warning signal and warning execution instruction are generated according to the danger level of the obstacle; the warning execution instruction is used to instruct the automobile warning component to work according to the warning signal.

[0041] In a third aspect, the present application also provides a computer device including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the steps of any of the above-mentioned edge computing-based automobile obstacle warning methods.

[0042] In a fourth aspect, the present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned edge computing-based automobile obstacle warning methods are implemented.

[0043] This edge computing-based automotive obstacle warning system uses a unified data acquisition module to comprehensively collect radar, image, and environmental parameters, significantly improving the system's detection capabilities in adverse weather, complex lighting, or obstructed conditions. By placing the core computing logic on the vehicle's edge computing platform, the recognition and response chain is effectively shortened, improving the vehicle's reliability and independent operation capabilities in disconnected or weak network environments, as well as its warning response speed. The combined output of obstacle type and danger level provides structural support for the subsequent grading and customization of response strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0045] Figure 1 This is a structural diagram of the edge computing-based automobile obstacle warning system of the present invention;

[0046] Figure 2 This is a schematic diagram of the multi-source data fusion process of the automobile obstacle warning system based on edge computing of the present invention;

[0047] Figure 3 This is a flow chart of the edge computing-based automobile obstacle warning method of the present invention. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0049] In one embodiment, Figure 1 As shown, an edge computing-based obstacle warning system for automobiles is provided. This embodiment uses the system applied to automobile components as an example. The new millimeter-wave radar is located in the center of the inner side of the front bumper of the vehicle, covering a 120-degree range in front. The high-definition multispectral camera is installed above the rearview mirror near the windshield, covering a 150-degree field of view in front. The ambient light sensor is located on the vehicle dashboard near the front windshield. The temperature and humidity sensors are located on the side of the vehicle body near the air intake. The edge computing processing unit is located inside the vehicle center console and is installed in the processing chip and storage module. The warning light is represented by a triangular luminous icon on the dashboard. The buzzer is the in-car audio system. The seat vibration motor is in the driver's seat. In this embodiment, the system includes:

[0050] The system includes a data acquisition module 110, an edge computing module 120 and an early warning module 130 which are connected in sequence via a data transmission bus.

[0051] In the system's overall architecture, communication and data exchange between functional modules rely on a highly reliable, high-bandwidth data transmission bus. This bus runs throughout the system, enabling the orderly connection and real-time coordination of the data acquisition module, edge computing module, and early warning module. This bus serves as a data carrier and ensures efficient and stable system response. Schematically, the data transmission bus utilizes a high-speed differential signal channel design, offering low latency and strong interference immunity, enabling stable data transmission even in complex onboard electromagnetic environments. After completing initial multi-source data acquisition and signal conditioning, the data acquisition module first encapsulates obstacle physical parameters, multispectral image data, and environmental factors into multimodal data frames, which are then transmitted to the edge computing module via the bus in a frame-structured format. Upon receiving the complete data frame, the edge computing module directly parses and processes it without requiring a central gateway or cloud-based transfer. This end-to-end data transmission path significantly reduces processing latency, improving the real-time performance and controllability of the entire system. After the calculations are complete, the edge computing module generates a warning signal packet based on the recognition results and risk level, and simultaneously transmits it to the early warning module via the same bus. This process ensures a closed loop from perception, decision-making to execution, meeting the system's requirements for data synchronization and response stability.

[0052] Optionally, the data transmission bus supports multiple industrial standard protocols such as CAN-FD and Ethernet AVB, and can be flexibly configured according to vehicle platform differences, further enhancing the system's adaptability and portability between different vehicle models.

[0053] The data acquisition module 110 is used to acquire multi-source data of the vehicle's surrounding environment collected by the sensor module; the multi-source data includes obstacle data, multispectral image data and environmental data.

[0054] The data acquisition module is primarily responsible for acquiring and aggregating multi-source environmental data about the vehicle's surroundings collected by sensor modules. For example, this multi-source data is acquired through a new millimeter-wave radar, a high-definition multispectral camera, and environmental perception sensors, including ambient light sensors and temperature and humidity sensors.

[0055] The millimeter-wave radar utilizes frequency-modulated continuous wave (FMCW) technology, continuously emitting electromagnetic waves within the 76 to 81 GHz operating frequency band. By receiving reflected signals, it analyzes the relative distance, velocity, and azimuth between the vehicle and the obstacle. Distance measurement accuracy reaches ±0.1 meter, velocity measurement accuracy reaches ±0.1 m / s, and angle measurement accuracy reaches ±1°. FMCW technology enables stable operation even in adverse weather conditions with low visibility, significantly improving target detection robustness. A high-definition multispectral camera, located in front of the vehicle or in the rearview mirror, captures visible and infrared image information, enabling clear identification of object outlines and thermal radiation signatures even in complex lighting conditions. Furthermore, the camera supports autofocus and image stabilization via commands from the edge technology module, ensuring image data stability and continuity. An ambient light sensor and a temperature and humidity sensor collect real-time ambient light intensity, temperature, and humidity parameters, providing external reference signals for subsequent image and radar data fusion.

[0056] The edge computing module 120 is used to perform fusion calculations on multi-source data and identify obstacles based on the obstacle recognition model to obtain the obstacle type and the corresponding obstacle danger level.

[0057] Schematically, the edge computing module relies on on-board high-performance computing platforms such as Drive AGX Xavier, and uses parallel neural network acceleration units to perform fusion calculations, feature extraction, and obstacle recognition on multi-source data. Schematically, data from the data acquisition module is aligned according to timestamps and then deeply fused within a unified embedded feature space. For example, a fusion recognition model based on a dual-branch convolutional neural network structure is used. Specifically, the radar subnetwork extracts dynamic feature maps such as speed trends and spatial distance changes of obstacles through a one-dimensional convolutional layer; the visual subnetwork uses multi-layer convolution and attention mechanisms to extract semantic information such as shape, color, and edge contours from the image. The intermediate outputs of the two networks are fused using a cross-modal attention mechanism to generate a joint feature vector with strong consistency and low redundancy.

[0058] The fusion results are then fed into the obstacle recognition module, which, combined with a classifier, outputs an obstacle type label. The risk assessment module then calculates the risk level based on the obstacle's type, location, relative speed, dynamic trends, and environmental parameters. To quantify this risk level, a multi-factor scoring function is introduced, integrating the obstacle type risk weight, dynamic approach rate, path overlap probability, and environmental perturbation factor. This output is a continuous risk score, which is then categorized into low, medium, and high levels for subsequent warning decisions.

[0059] The warning module 130 is used to generate corresponding warning signals and warning execution instructions according to the danger level of the obstacle; the warning execution instructions are used to instruct the automobile warning components to work according to the warning signals.

[0060] The warning module converts the obstacle danger level signal output by the edge computing module into specific warning instructions, and drives the warning components within the vehicle to execute multimodal prompts. For example, when the recognition result indicates a high risk level, the system will trigger the highest level of warning, including a rapid flashing instrument panel warning light, a high-frequency beeping sound reminder, and strong vibration feedback from the driver's seat vibrator. If the risk level is medium, the system will appropriately combine audio and visual warnings with mild vibrations. If the risk level is low, the system will only alert the driver through visual prompts or recordings. In addition, all warning signals are recorded with timestamps and can be linked to the vehicle control system to provide auxiliary decision-making basis for intelligent driving behavior.

[0061] In the aforementioned edge computing-based automotive obstacle warning system, the data acquisition module simultaneously collects multimodal information from sensor modules, including obstacle data, multispectral image data, and environmental data. This effectively integrates multi-source environmental data, ensuring the system acquires comprehensive and rich environmental perception information, providing multi-angle support for subsequent analysis. By deploying the computational processing unit locally within the vehicle's edge computing module, the system performs data fusion and obstacle identification locally in real time, avoiding the communication delays associated with traditional cloud-based servers. This significantly improves overall warning response speed, making it particularly suitable for emergency response requirements during high-speed driving or in situations involving sudden obstacles. The edge computing module not only identifies obstacle type but also further determines the degree of danger by combining obstacle status and location. This provides the system with automated risk assessment capabilities, provides a quantitative basis for subsequent response, and reduces the possibility of human intervention and misjudgment. Based on the recognition results, the warning module generates warning signals and warning execution instructions, controlling the vehicle's warning components to operate in different modes, thereby implementing a multi-channel warning method using vision, hearing, and touch, enhancing the driver's awareness of potential hazards.

[0062] In one embodiment, the data acquisition module includes a millimeter wave radar sensing unit, a multispectral camera driving unit and a sensor module unit.

[0063] The millimeter-wave radar sensing unit is used to calculate the time difference and frequency change between the transmitted wave and the reflected wave through frequency-modulated continuous waves to obtain obstacle data; obstacle data includes distance data, speed data and angle data.

[0064] Schematically, the millimeter-wave radar sensing unit uses a radar module based on frequency-modulated continuous wave (FMCW) technology. Specifically, the radar continuously transmits linear frequency-modulated millimeter-wave signals and synchronously receives signals reflected by obstacles in front. By comparing the time delay and frequency difference between the transmitted and reflected waves, the relative distance, speed, and angle of incidence between the obstacle and the vehicle are calculated. For example, the distance is calculated based on the wave propagation time difference formula, the speed is obtained through Doppler shift analysis, and the angle is obtained by measuring the phase difference between multiple array receiving antennas.

[0065] The multispectral camera driving unit is used to drive the multispectral camera to collect visible light images and infrared light images of the vehicle's surrounding environment at a preset frame rate, and preprocess the obtained visible light images and infrared light images to obtain multispectral image data; the corresponding preprocessing steps include denoising and enhancement.

[0066] The multispectral camera driver unit is responsible for acquiring image information of the vehicle's surroundings and improving image quality to enhance the robustness of visual recognition. Specifically, a multispectral camera array that senses visible light and infrared light synchronously captures environmental images at a system-set frame rate, such as 30 frames per second, with an image resolution of 1920×1080 pixels. The collected raw image data then enters the image preprocessing process. For example, denoising is performed using methods such as Gaussian filtering and wavelet noise reduction to eliminate image interference caused by low light, dust, raindrops, etc.; enhancement processing improves image contrast, edge contours, or hot zone details, making it easier for subsequent image analysis modules to identify the structure and status of obstacles. The processed image is output as multispectral image data composed of a combination of visible light and infrared images, fully covering environmental perception needs under different lighting and weather conditions.

[0067] The sensor module unit is used to perform standardized numerical conversion processing on the analog signals of external light intensity value, temperature value and humidity collected by the sensor module to obtain environmental data.

[0068] The sensing module unit is responsible for collecting and converting data related to the environmental status. The analog electrical signals of the ambient light sensor and the temperature and humidity sensor, which obtain the external light intensity, air temperature and relative humidity values ​​in real time, are processed by the signal conditioning and standardization module. Among them, the working light range of the ambient light sensor is 0-100000 Lux, the temperature and humidity sensor measures the temperature range from -50°C to 125°C, and the humidity measurement range is 0-100% RH. Schematically, the analog signal is converted into a digital signal through an analog-to-digital conversion (ADC) circuit, and then the raw reading is converted into a standardized environmental data value through normalization or unit conversion. For example, the light intensity is in Lux, the temperature is expressed in degrees Celsius, and the humidity is expressed as a percentage RH. The standardized environmental data will be used as an environmental reference item in the multimodal data and transmitted to the edge computing module to assist in determining the recognizability of the obstacle image and the dynamic adjustment of the recognition strategy.

[0069] In one embodiment, the edge computing module includes a data fusion unit and an obstacle recognition unit.

[0070] The data fusion unit is used to fuse the obstacle data and the multispectral data through a data fusion algorithm, and optimize the obstacle data with the environmental data to obtain obstacle environment fusion data.

[0071] The data fusion unit fuses obstacle data from the millimeter-wave radar sensing unit with image data provided by the multispectral camera driver unit at the feature level within a unified spatiotemporal domain. Schematically, the fusion algorithm is based on temporal and spatial alignment mechanisms. In the temporal dimension, radar data frames and image frames are synchronized using timestamps. In the spatial dimension, coordinate mapping is used to project the radar detection area onto the image coordinate system. This allows for high-precision alignment of the range, velocity, and angle information returned by the radar with the spatial pixels and thermal image profiles in the image. Furthermore, an attention-based data fusion architecture is introduced, adaptively adjusting the feature fusion strength based on the signal-to-noise ratio of different modalities in the current environment, thereby enhancing the expressiveness of key obstacle features and mitigating environmental interference. The system also incorporates environmental data as optimization criteria, specifically parameters such as light intensity, temperature, and humidity, which influence the weighting process of data fusion and subsequent feature enhancement. For example, in low-light environments, the system adaptively increases the weight of the infrared channel image and reduces the contribution of image texture features in the fusion output. In high-humidity environments, an image fog reduction algorithm or thermal feature smoothing strategy can be introduced to improve recognition robustness. Finally, the fusion result is output in the form of obstacle environment fusion data and sent to the obstacle recognition unit as an input signal.

[0072] The obstacle recognition unit is used to input the obstacle environment fusion data into the pre-trained obstacle recognition model to obtain the obstacle type, and determine the obstacle danger level based on the vehicle's driving status, combined with the obstacle type and the obstacle's position and speed data; the obstacle's position is determined by distance data and angle data.

[0073] The obstacle recognition unit performs high-level semantic analysis based on the fused data. It employs a pre-trained deep learning model for obstacle recognition. Schematically, this model is based on a convolutional neural network (CNN) architecture and trained on a large number of multi-source obstacle samples in complex road scenarios. It possesses high-precision classification capabilities for common obstacle types, including pedestrians, two-wheeled vehicles, small and large vehicles, roadblocks, and floating objects. Specifically, the obstacle recognition unit inputs the fused data into the model, rapidly outputs obstacle type labels, and further extracts information about the obstacle's motion, such as its position and velocity.

[0074] Optionally, the obstacle's position is determined using a combination of distance and angle data provided by the radar. Polar coordinate transformation converts the two-dimensional angle-distance information into positional information in a standard rectangular coordinate system, enabling precise positioning of the obstacle relative to the vehicle's reference frame. The system also incorporates real-time vehicle status data, including current speed, acceleration, and lane keeping status, and constructs a comprehensive evaluation function to determine the current obstacle's danger level based on indicators such as obstacle type, relative speed, relative position, and vehicle status.

[0075] The calculation of the degree of danger depends not only on static indicators such as the type of obstacle and whether it is in front of the lane, but also on dynamic states such as whether the obstacle is approaching, its speed and relative direction, and environmental constraints, that is, whether it can be avoided in time. The final output is a continuous risk score or a discrete level label, which is used to drive the warning module to implement a response strategy of the corresponding level.

[0076] The edge computing module of the above system realizes the complete link processing from the underlying perception data to the advanced intelligent decision-making, enabling the entire system to have the ability of rapid response, high-precision identification and adaptation to complex environments, providing a solid intelligent foundation and computing support for the realization of truly practical intelligent assisted driving obstacle warning functions.

[0077] In one embodiment, Figure 2 As shown in the figure, obstacle data and multispectral data are fused through data fusion algorithm and optimized with environmental data to obtain obstacle environment fusion data, including:

[0078] S201. Obtain temporal speed-distance graph features from the distance data, speed data, and angle data of the obstacle data through multiple convolutional layer features.

[0079] Schematically, a temporal neural network structure composed of multiple one-dimensional convolutional layers is used to extract features from obstacle data, including distance, speed, and angle data. The data is then gradually extracted, gradually analyzing the speed variation patterns and spatial distribution trends over time. Through continuous convolution and pooling operations, the network constructs a speed-distance map with strong temporal perception, revealing the relative motion trajectory of obstacles and providing a dynamic support dimension for subsequent fusion with image structural features.

[0080] S202: Extract structural contours from multispectral image data using ResNet50 and an asymmetric attention module to obtain contour features.

[0081] Schematically, the residual network structure ResNet50 is used as the backbone network. With its deep convolutional channels and skip connection characteristics, it effectively enhances the ability to extract image texture, edges, and hot zone contours for visible light and infrared images, capturing detailed information and thermal radiation characteristics under lighting conditions. At the same time, to further improve the expression adaptability of image features in modal fusion, an asymmetric attention mechanism is introduced to dynamically adjust the attention weights of each image region, so that key obstacle edges and hot spots receive a higher response in the feature map, resulting in a set of high-dimensional structural contour features that represent the spatial semantics and thermal distribution information contained in the image data.

[0082] S203. Use the cross-modal attention mechanism to unify the temporal speed-distance map features and contour features, and optimize them in combination with the environmental data to obtain obstacle environment fusion data.

[0083] Schematically, to address the heterogeneity of different data sources in terms of expression form, feature scale, and semantic level, the system constructs a fusion architecture based on a cross-modal attention mechanism. This mechanism establishes an aligned attention map between modalities, enabling efficient matching and interaction between temporal velocity-distance map features and structural contour features in a unified embedding space, thereby forming a highly coupled joint expression while maintaining the advantages of each feature. During the fusion process, the system introduces environmental data as an optimization factor, where light, temperature, and humidity information are used to dynamically adjust the fusion weights of each modal feature. The output is a set of obstacle environment fusion data that combines the dynamic state of obstacles with visual structural information and environmental perception capabilities.

[0084] In one embodiment, the obstacle recognition model obtains the obstacle type by the following method:

[0085] S31. Extract the physical form, dynamic state and risk situation of the obstacle based on the obstacle environment fusion data.

[0086] Entity morphological feature extraction mainly relies on the structural contour features extracted by ResNet50 and the attention mechanism in the image feature channel. Combined with the size range information such as the target scattering intensity and distance diffusion width provided by the millimeter-wave radar, the physical manifestation of obstacles in terms of shape, size, edge complexity, etc. are judged to preliminarily distinguish between different forms of physical objects such as pedestrians, bicycles, motor vehicles, and roadblocks.

[0087] The dynamic state uses dynamic signals such as the speed gradient and acceleration change trend in the aforementioned time-series speed-distance map, combined with the vehicle's own speed vector, to determine whether the obstacle is stationary, moving slowly, or approaching quickly through relative motion calculation.

[0088] Risk situation analysis combines the obstacle location coordinates with the vehicle's current path, lane model, and driving behavior status such as whether to change lanes or turn, and performs spatial comparison to determine whether the obstacle is on the driving path, whether it blocks the driving field of view, or whether it is in an uncontrollable area.

[0089] S32. Classify the category labels according to the entity shape, dynamic state and risk situation to obtain the obstacle type.

[0090] The entity's form, dynamic state, and risk profile are integrated by building a multi-label classifier or label mapping rules to output a comprehensive category label for the obstacle. Examples of obstacle types include pedestrians approaching quickly and in the center of the path, small vehicles moving slowly and in an avoidable zone, or static roadblocks with high occlusion and blind spots.

[0091] In one embodiment, the obstacle danger level is obtained by:

[0092] S41. Based on the driving state of the vehicle, score the obstacle type to obtain a danger score.

[0093] The risk score is obtained using the following formula:

[0094] Risk Score=α·T r ·V r ·P r ·E r

[0095]

[0096] Among them, RiskScore is the risk score; T r is the obstacle type risk factor; V r is a dynamic predictor; P r Position information factor; E r is the environmental disturbance factor; α is the adjustment factor; k is the speed control growth rate; V relis the relative speed between the obstacle and the car; v th is the dangerous speed threshold.

[0097] The danger score is a numerical representation of the threat level of obstacles to safety in the current driving state, which enables the system to automatically determine whether to trigger an early warning and what level of early warning response to trigger. r Obstacle type risk factors can be assigned according to the potential threat level of different types of obstacles. For example, pedestrians are set to 0.9, stationary roadblocks are set to 0.6, and floating objects are set to 0.3; V r Dynamic prediction factor, which characterizes the risk intensity caused by the relative speed of the obstacle; P r Position information factor, reflecting whether the obstacle is in the current driving path of the car. For example, if it is in the center of the path, it is assigned a value of 1, if it is in the avoidable area, it is assigned a value of 0.5, and if it is outside the path, it can be assigned a value of 0.3; E r Environmental disturbance factor, which determines the perception stability coefficient based on the current ambient light, temperature, humidity and other information. For example, under conditions such as dense fog, high humidity, and low light, E r The value is appropriately lowered to 0.7-0.8 to improve the model's sensitivity to environmental interference; the α global adjustment factor is used to control the distribution range of the overall score to ensure the separability of distributions at different levels and system stability.

[0098] S42. Classify the risk scores according to the preset threshold interval to obtain the risk level of the obstacle; the risk level of the obstacle includes low risk, medium risk and high risk.

[0099] Schematically, the risk score is classified into different levels according to preset threshold intervals. The classification mechanism maps the continuous value Risk Score to a discrete risk level label through the threshold boundary. For example, if the Risk Score is greater than 0.7, it is judged as high risk, indicating that the obstacle has a high probability of collision or seriously affects driving behavior, and the system will trigger the strongest level of warning response; if 0.4 < Risk Score < 0.7, it is judged as medium risk, alerting the driver to potential threats ahead, with warnings mainly in the form of sound and light prompts; if the Risk Score is less than 0.4, it is judged as low risk, and only visual annotation or background recording is performed, without interfering with driving behavior.

[0100] Optionally, the hierarchical strategy is highly real-time and flexible. The system can dynamically update the threshold strategy based on the actual driving scenario, such as relaxing the medium and high risk boundaries when driving at low speeds in urban areas, and tightening the threshold conditions when driving on highways, thereby improving the adaptability and accuracy of the overall warning system.

[0101] In one embodiment, the early warning module includes an early warning signal generating unit and an early warning executing unit;

[0102] The warning signal generation unit is used to generate corresponding warning signals according to the preset warning strategy based on the danger level of the obstacle; the warning signals include light flashing signals, sound signals and seat vibration signals;

[0103] The warning execution unit is used to generate multi-dimensional warning execution instructions based on the warning signal; the warning execution instructions include a light instruction for instructing an indicator light of a control panel to flash according to a light flashing signal, a sound instruction for instructing a power amplifier circuit to drive a buzzer according to a sound signal, and a vibration instruction for instructing a pulse width modulation to drive a vibration electrode to vibrate according to a seat vibration signal;

[0104] Among them, the preset early warning strategy corresponds to the following steps:

[0105] When the obstacle danger level is medium risk, the warning signal includes a light flashing signal at a preset low frequency threshold, a sound signal at a preset medium threshold intensity, and a seat vibration signal at a preset light threshold intensity.

[0106] When the obstacle danger level is high risk, the warning signal includes a light flashing signal of a preset high frequency threshold, a sound signal of a preset high threshold intensity, and a seat vibration signal of a preset strong threshold intensity.

[0107] Schematically, the warning module is used to communicate potential safety risks to the driver. The warning signal generation unit, based on the danger level output by the obstacle recognition module, calls upon the system's pre-set warning strategy rule library to generate a corresponding warning signal combination. Warning signals are expressed in three dimensions: visual, auditory, and tactile. Specifically, they include: a light flashing signal that controls the instrument panel warning lights to flash at different frequencies to attract visual attention; an audio signal that controls the buzzer to output beeps of varying intensities and tones to enhance the acoustic warning effect; and a seat vibration signal that generates vibrations of varying frequencies and amplitudes by driving the seat's internal vibration motor, delivering tactile cues.

[0108] In terms of signal generation strategy, the system sets differentiated response parameters for different risk levels. For example, when the obstacle danger level is identified as medium risk, the system generates the following warning signal combination: the light flashing signal is set to 1Hz flashing, the sound signal is 60-70dB beeping, and the seat vibration signal is a low-rate short-term vibration. This strategy is intended to alert the driver rather than force intervention.

[0109] When the danger level of an obstacle is judged to be high risk, the system immediately generates a high-intensity warning signal: the light flashing signal is set to 3-5Hz, the sound signal output is >85dB, and a continuous strong vibration seat vibration signal is activated simultaneously, aiming to intervene in the driver's perception through multiple channels in a short period of time and strive for the most timely attention response.

[0110] The above signals are encapsulated into instruction packets in a standardized format by the generation unit and transmitted in real time to the warning execution unit. This unit is responsible for converting the signals into electrical control instructions that can directly drive the actuators in the vehicle and accurately assigning them to the corresponding execution path. Specifically, the light instruction is used to control the warning lights on the instrument panel or HUD in the vehicle. According to the frequency setting of the light flashing signal, the LED flashing rhythm is controlled by the signal modulation circuit; the sound instruction is used to instruct the power amplifier circuit of the buzzer system to control the voltage / current according to the intensity of the sound signal, thereby driving the buzzer to emit a sound of the corresponding decibel and frequency; the vibration instruction controls the vibration motor installed under the driver's seat through the pulse width modulation module (PWM), generating short or continuous, mild or strong vibration feedback according to the signal instruction.

[0111] The hardware design of this early warning execution architecture ensures a response time of less than 100 milliseconds, ensuring the system's full responsiveness in high-risk emergencies. Furthermore, the multi-dimensional signal channels are not logically equivalent but rather combinable and redundant. For example, in situations where the driver's vision is limited, the system prioritizes sound and vibration signals to enhance the effectiveness of warnings, demonstrating its intelligent integration of scene perception and execution strategies.

[0112] The system was installed on an experimental vehicle and placed in a large, closed test field capable of simulating a variety of road conditions. Various types of obstacles were set up, and the test vehicles were driven at various speeds, including 30 km / h, 60 km / h, and 90 km / h. Testing of the obstacles revealed that the new millimeter-wave radar and high-definition multispectral camera worked in tandem, effectively compensating for each other's shortcomings. In complex weather conditions, such as heavy rain and dense fog, the detection accuracy of traditional millimeter-wave radars decreased significantly, with distance errors reaching ±2 meters. However, the distance measurement accuracy of the system presented in this invention remained stable at ±0.1 meters, achieving an accuracy improvement of 95%. In low-light and complex background scenes, traditional cameras are prone to misjudgments and missed detections. The high-definition multispectral camera presented in this invention, combined with a novel algorithm, increased obstacle recognition accuracy from 70% to over 95%, significantly improving detection accuracy. The use of an edge computing processing unit significantly reduces data processing latency. When the vehicle is traveling at 100 km / h, traditional technology delays the sensor data acquisition and warning by approximately 0.5 seconds, leaving the driver with minimal reaction time. The system of the present invention can reduce the warning delay to less than 0.1 seconds, 0.4 seconds in advance, which can buy more valuable reaction time for the driver in emergency situations and effectively reduce the risk of accidents. By collecting data through environmental sensors and combining specific algorithms, the system can adapt to a variety of complex scenarios. On unpaved roads, traditional systems have poor detection effects on irregular obstacles, while the detection accuracy of the system of the present invention can reach more than 85%; in extreme weather scenarios with drastic changes, such as rapid cooling accompanied by road icing, the system of the present invention can adjust the detection strategy in time according to environmental parameters to maintain stable performance. Compared with traditional technologies that can only work in conventional scenarios, the applicable scenarios have increased by more than 50%, greatly improving the safety of cars driving in various road conditions and weather conditions.

[0113] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0114] Based on the same inventive concept, the embodiments of the present application also provide an edge computing-based automobile obstacle warning method for implementing the aforementioned edge computing-based automobile obstacle warning system. The implementation solution provided by this method is similar to the implementation solution described in the aforementioned method. Therefore, the specific limitations of one or more embodiments of the edge computing-based automobile obstacle warning method provided below can be found in the above-mentioned limitations of the edge computing-based automobile obstacle warning system, and will not be repeated here.

[0115] In an exemplary embodiment, Figure 3 As shown, a vehicle obstacle warning method based on edge computing is provided, including:

[0116] S301. Acquire multi-source data of the vehicle's surrounding environment collected by a sensor module; the multi-source data includes obstacle data, multispectral image data, and environmental data.

[0117] S302: Perform fusion calculation on multi-source data, and identify obstacles according to the obstacle identification model to obtain obstacle types and corresponding obstacle danger levels.

[0118] S303. Generate a corresponding warning signal and warning execution instruction according to the danger level of the obstacle; the warning execution instruction is used to instruct the automobile warning component to work according to the warning signal.

[0119] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0120] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0121] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the components described as separate parts may or may not be physically separated, and the parts displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0122] The above-described embodiments merely represent several implementation methods of the embodiments of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the concept of the embodiments of the present application, and these modifications and improvements fall within the scope of protection of the embodiments of the present application.

Claims

1. An automobile obstacle warning system based on edge computing, characterized by: The system includes a data acquisition module, an edge computing module and an early warning module connected in sequence via a data transmission bus; The data acquisition module is used to acquire multi-source data of the vehicle's surrounding environment collected by the sensor module; the multi-source data includes obstacle data, multispectral image data and environmental data; The edge computing module is used to perform fusion calculation on the multi-source data and identify obstacles according to the obstacle recognition model to obtain the obstacle type and the corresponding obstacle danger level; The warning module is used to generate a corresponding warning signal and a warning execution instruction according to the danger level of the obstacle; the warning execution instruction is used to instruct the automobile warning component to work according to the warning signal.

2. The system according to claim 1, wherein: The data acquisition module includes a millimeter wave radar perception unit, a multispectral camera driving unit and a sensor module unit; The millimeter wave radar sensing unit is used to calculate the time difference and frequency change between the transmitted wave and the reflected wave by frequency modulating the continuous wave to obtain obstacle data; The obstacle data includes distance data, speed data and angle data; The multispectral camera driving unit is used to drive the multispectral camera to collect visible light images and infrared light images of the vehicle's surrounding environment at a preset frame rate, and pre-process the obtained visible light images and infrared light images to obtain the multispectral image data; The pre-processing corresponding steps include denoising and enhancement; The sensor module unit is used to perform standardized numerical conversion processing on the analog signals of the external light intensity value, temperature value and humidity collected by the sensor module to obtain environmental data.

3. The system according to claim 2, characterized in that: The edge computing module includes a data fusion unit and an obstacle recognition unit; The data fusion unit is used to fuse the obstacle data and the multispectral data through a data fusion algorithm, and optimize the obstacle environment data to obtain obstacle environment fusion data; The obstacle recognition unit is configured to input the obstacle environment fusion data into the pre-trained obstacle recognition model to obtain an obstacle type, and determine the obstacle hazard level based on the vehicle's driving state, the obstacle type, the obstacle's position, and the speed data; the obstacle's position is determined by the distance data and the angle data.

4. The system according to claim 3, characterized in that The step of fusing the obstacle data and the multispectral data by a data fusion algorithm and optimizing the environment data to obtain obstacle environment fusion data comprises: The distance data, the speed data and the angle data of the obstacle data are processed through multiple convolutional layer features to obtain a time series speed-distance graph feature; Performing structural contour extraction on the multispectral image data using ResNet50 and an asymmetric attention module to obtain contour features; The temporal speed-distance graph features and the contour features are unified using a cross-modal attention mechanism, and are optimized in combination with the environmental data to obtain the obstacle environment fusion data.

5. The system according to claim 3, wherein: The obstacle recognition model obtains the obstacle type by the following method: Extracting the entity form, dynamic state and risk situation of the obstacle according to the obstacle environment fusion data; Classification of category labels is performed according to the entity form, dynamic state and risk situation to obtain the obstacle type.

6. The system according to claim 5, characterized in that The obstacle danger level is obtained by: Based on the driving state of the vehicle, scoring is performed according to the obstacle type to obtain a danger score; Classifying the hazard scores according to a preset threshold range to obtain the hazard level of the obstacle; The obstacle hazard levels include low risk, medium risk and high risk; The risk score is obtained by the following formula: Risk Score=α·T r ·V r ·P r ·E r Wherein, RiskScore is the risk score; T r is the obstacle type risk factor; V r is a dynamic predictor; P r Position information factor; E r is the environmental disturbance factor; α is the adjustment factor; k is the speed control growth rate; V rel is the relative speed between the obstacle and the car; v th is the dangerous speed threshold.

7. The system according to claim 6, characterized in that: The early warning module includes an early warning signal generating unit and an early warning execution unit; The warning signal generating unit is used to generate a corresponding warning signal according to the danger level of the obstacle according to a preset warning strategy; the warning signal includes a light flashing signal, a sound signal and a seat vibration signal; The warning execution unit is used to generate multi-dimensional warning execution instructions based on the warning signal; the warning execution instructions include a light instruction for instructing an indicator light of the control panel to flash according to the light flashing signal, a sound instruction for instructing a power amplifier circuit to drive a buzzer according to the sound signal, and a vibration instruction for instructing a pulse width modulation to drive a vibration electrode to vibrate according to the seat vibration signal; The preset warning strategy corresponds to the following steps: When the obstacle hazard level is medium risk, the warning signal includes a light flashing signal at a preset low frequency threshold, a sound signal at a preset medium threshold intensity, and a seat vibration signal at a preset light threshold intensity; When the obstacle danger level is high risk, the warning signal includes a light flashing signal at a preset high frequency threshold, a sound signal at a preset high threshold intensity, and a seat vibration signal at a preset strong threshold intensity.

8. A vehicle obstacle warning method based on edge computing, characterized in that: The method comprises: Acquire multi-source data of the vehicle's surrounding environment collected by the sensor module; the multi-source data includes obstacle data, multispectral image data, and environmental data; Performing fusion calculation on the multi-source data and performing obstacle recognition according to an obstacle recognition model to obtain obstacle types and corresponding obstacle danger levels; A corresponding warning signal and warning execution instruction are generated according to the danger level of the obstacle; the warning execution instruction is used to instruct the automobile warning component to work according to the warning signal.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to claim 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to claim 8 are implemented.

Images