Intelligent automobile roadblock recognition system and method based on millimeter waves, and automobile

Abstract

The invention relates to an intelligent automobile roadblock recognition system and method based on millimeter waves and an automobile, and the system is characterized in that the system comprises millimeter wave radar sensors which are installed at the front end and the side of the intelligent automobile; the signal processing module is connected with the millimeter wave radar sensor, and the feature extraction module is connected with the signal processing module; the roadblock classification module is connected with the feature extraction module; the path planning module is connected with the roadblock classification module and re-plans the driving path of the intelligent automobile in real time according to the type and the position of the roadblock and the current position and the driving direction of the intelligent automobile so as to avoid the roadblock and ensure the driving safety; and the control module is connected with the path planning module and used for generating a corresponding control instruction according to the planned driving path and controlling the driving direction, speed and the like of the intelligent automobile so that the intelligent automobile can drive according to the planned path.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent vehicles, and in particular to a millimeter wave-based intelligent vehicle roadblock recognition system and method, as well as a vehicle. Background Art

[0002] With the rapid development of the smart car industry, roadblock recognition, as a key environmental perception technology, directly impacts the safety and reliability of autonomous and assisted driving. In complex traffic environments, smart cars must quickly and accurately identify various roadblocks—fixed obstacles, moving vehicles, pedestrians, and traffic facilities—to make informed decisions and plan routes to ensure safe driving.

[0003] Traditional roadblock recognition technologies primarily rely on cameras, LiDAR, and ultrasonic sensors. Cameras capture rich image information, but are susceptible to changes in lighting and weather conditions, and have limited depth perception. LiDAR provides high-precision 3D point cloud data, but is costly and has limited performance in complex weather conditions. Ultrasonic sensors can effectively detect obstacles at short distances, but their limited detection range makes them difficult to identify roadblocks at long distances.

[0004] Millimeter-wave radar is gaining attention due to its unique advantages. It can transmit and receive millimeter-wave signals, typically in the 76-81 GHz frequency range. It operates in all weather conditions and can penetrate rain, fog, and dust. Its compact size, high resolution, and strong anti-interference capabilities allow it to measure target distance, speed, and angle in real time, making it suitable for identifying obstacles in complex environments.

[0005] However, existing millimeter-wave radar applications in roadblock recognition still have some shortcomings. First, single millimeter-wave radar systems have limited target classification accuracy and are unable to distinguish specific roadblock types. Second, data fusion technology between different sensors needs to be improved. Sensor data exhibits modal differences. For example, millimeter-wave radar provides target distance, speed, and angle information, while cameras provide rich visual features. Efficiently fusing this heterogeneous data to improve the accuracy and reliability of roadblock recognition is a technical challenge. Furthermore, real-time and stability are key challenges in roadblock recognition. Smart cars traveling at high speeds require roadblock recognition systems to complete data collection, processing, and decision-making within a short period of time, and must maintain stable and reliable operation under complex road conditions and changing environments. Furthermore, existing roadblock recognition technology is relatively weak in hazard assessment and graded warnings, making it difficult to accurately quantify the hazard level of roadblocks and implement targeted warning measures. Summary of the Invention

[0006] The present invention proposes an intelligent automobile roadblock recognition system based on millimeter waves, which solves the above-mentioned problems existing in the use process of the prior art.

[0007] The technical solution of the present invention is achieved as follows:

[0008] An intelligent automobile roadblock recognition system based on millimeter waves, characterized by comprising:

[0009] Millimeter-wave radar sensors, installed on the front and sides of smart cars, are used to transmit millimeter-wave signals and receive reflected signals;

[0010] The signal processing module is connected to the millimeter wave radar sensor and is used to pre-process the reflected signal, including filtering, amplification and analog-to-digital conversion operations, to extract the original feature information of the target object;

[0011] The feature extraction module is connected to the signal processing module and is used to extract various features of the target object from the pre-processed reflection signal, including distance, speed, angle, radar cross-section area and target shape features;

[0012] The roadblock classification module, connected to the feature extraction module, is built based on a machine learning algorithm. A classification model is pre-trained using a dataset containing sample features of different types of roadblocks. The module is used to classify and identify target objects based on the extracted target features, determining whether they are roadblocks and the specific type of roadblock. Roadblock types include fixed obstacles, moving vehicles, pedestrians, and traffic facilities.

[0013] The path planning module is connected to the roadblock classification module. It replans the smart car's driving path in real time based on the type and location of the roadblock and the current position and driving direction of the smart car to avoid the roadblock and ensure driving safety.

[0014] The control module is connected to the path planning module and is used to generate corresponding control instructions according to the planned driving path, control the driving direction and speed of the smart car, and make the smart car drive according to the planned path.

[0015] Preferably, the millimeter-wave radar sensor includes multiple transmitting antennas and multiple receiving antennas to form a multi-input multi-output radar array, which can improve the resolution and accuracy of target detection and achieve simultaneous detection and tracking of multiple targets;

[0016] The signal processing module adopts an adaptive filtering algorithm, which can automatically adjust the filter parameters according to the changes in environmental noise, effectively suppress noise interference and improve signal quality;

[0017] The method for extracting target shape features by the feature extraction module includes a time-frequency analysis and spatial spectrum estimation algorithm based on millimeter wave signals, which can accurately extract the shape contour features of the target according to the differences in the reflection characteristics of different target objects to millimeter wave signals;

[0018] The roadblock classification module uses a convolutional neural network to build a classification model. The convolutional neural network classification model includes multiple convolutional layers, pooling layers and fully connected layers. By training on a large amount of sample data labeled with different types of roadblock features, it can automatically learn the feature representation of the target object and achieve high-precision classification and recognition of roadblock types.

[0019] Preferably, it also includes a data fusion module, which is respectively connected to the millimeter wave radar sensor, the camera of the smart car, the ultrasonic sensor and the vehicle communication system, and is used to fuse the data from different sensors and communication equipment, including Kalman filter fusion, particle filter fusion and other fusion algorithms to improve the accuracy and reliability of target detection and roadblock identification.

[0020] Preferably, the control module adopts a model predictive control algorithm to predict the future state of the smart car in real time based on the planned driving path and the dynamic model of the smart car, and generates corresponding control instructions, so that the smart car can maintain stable driving performance while avoiding road obstacles, thereby improving driving comfort.

[0021] A millimeter wave-based intelligent automobile roadblock recognition method includes the following steps: transmitting a millimeter wave signal and receiving a reflected signal through a millimeter wave radar sensor;

[0022] Preprocess the reflected signal to extract the original feature information of the target object;

[0023] Extracting multiple features of the target object from the preprocessed reflection signal;

[0024] Classify and identify the target object based on the extracted target features to determine whether it is a roadblock and the specific type of roadblock;

[0025] Replan the smart car's route in real time based on the type and location of the roadblock and the smart car's current location and direction of travel;

[0026] Generate corresponding control instructions based on the planned driving path to control the driving direction and speed of the smart car.

[0027] Preferably, before preprocessing the reflected signal, the method also includes the steps of time synchronization and coordinate conversion of the data collected by the millimeter wave radar sensor to ensure the consistency of the data collected by different sensors in time and space, thereby improving the accuracy of data fusion and target recognition.

[0028] Preferably, the extraction of multiple features of the target object includes calculating the speed of the target object based on the Doppler effect of the millimeter wave signal, estimating the angular position of the target object based on the signal arrival angle and departure angle, and calculating the distance and RCS characteristics of the target object through the amplitude and phase information of the signal.

[0029] Preferably, after classifying and identifying the target object, the method also includes calculating the danger level of the roadblock based on the classification results, and adjusting the alarm level and control strategy according to the danger level. When the danger level of the roadblock is high, the emergency braking system of the smart car is triggered in time or an audible and visual alarm prompt is issued to ensure driving safety.

[0030] Preferably, the corresponding control instructions are generated according to the planned driving path, including adjusting the steering angle and speed of the smart car according to the curvature and slope of the path, so that the smart car can maintain a reasonable driving speed and stable handling performance while avoiding road obstacles, thereby improving driving safety and comfort.

[0031] A smart car is equipped with the millimeter wave-based smart car roadblock recognition system.

[0032] In summary, the beneficial effects of the present invention are:

[0033] This invention discloses a millimeter-wave-based intelligent roadblock recognition system for vehicles. This millimeter-wave-based intelligent roadblock recognition technology demonstrates significant advantages and benefits in improving roadblock recognition accuracy, enhancing perception capabilities, optimizing path planning and control, accurately assessing hazard levels, and enhancing the driving experience and market competitiveness. The widespread application of this technology is expected to further promote the development and popularization of intelligent vehicle technology, laying a solid foundation for future intelligent transportation development. DETAILED DESCRIPTION

[0034] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0035] Example

[0036] This embodiment of the millimeter-wave-based intelligent vehicle roadblock recognition system is applied to a self-driving smart car, aiming to improve its driving safety and autonomous decision-making capabilities in complex road environments. It primarily consists of a millimeter-wave radar sensor, a signal processing module, a feature extraction module, a roadblock classification module, a path planning module, a control module, and a data fusion module.

[0037] Four millimeter-wave radar sensors are installed on the left and right sides of the front of the smart car, as well as on the front and rear sides. They operate in the 76-81 GHz frequency range. This high-frequency millimeter-wave signal has excellent directionality and resolution, enabling effective detection of targets in different directions, both in front and to the sides.

[0038] Each millimeter-wave radar sensor is equipped with four transmit antennas and four receive antennas, forming a multiple-input, multiple-output (MIMO) radar array. MIMO technology creates more virtual receive channels, significantly improving target detection resolution and accuracy. It enables simultaneous detection and tracking of multiple targets, accurately acquiring surrounding information even in complex road conditions.

[0039] (2) Signal processing module

[0040] The signal processing module is installed in the electronic control unit (ECU) of a smart car and is connected to the millimeter-wave radar sensor via a high-speed data bus. Its main function is to pre-process the reflected signals received by the radar sensor.

[0041] The module first filters the reflected signal using an adaptive filtering algorithm. This algorithm monitors changes in ambient noise in real time and automatically adjusts the filter parameters to effectively suppress various interference noises in the surrounding environment, such as electromagnetic interference from other electronic devices and clutter caused by reflections from road noise. The filtered signal is then amplified, and the signal amplitude is adjusted to a range suitable for subsequent analog-to-digital conversion, ensuring signal integrity and accuracy. Finally, a high-speed analog-to-digital converter converts the analog signal into a digital signal, providing the foundation for subsequent digital signal processing.

[0042] (3) Feature extraction module

[0043] The feature extraction module is also integrated into the ECU and is closely connected to the signal processing module. It is responsible for extracting various key features of the target object from the pre-processed digital signal.

[0044] To extract the distance signature of a target object, the module leverages the propagation time difference of millimeter-wave signals to accurately calculate the straight-line distance between the smart car and the target, with an error within centimeters. Velocity signature extraction leverages the Doppler effect. By analyzing the frequency offset of the reflected signal, the target's velocity relative to the smart car is calculated in real time, enabling precise differentiation between stationary obstacles and moving vehicles. Angle signature extraction utilizes angle-of-arrival (AOA) and angle-of-departure (DOA) estimation algorithms, combined with information from multiple receiving channels of the MIMO radar array, to determine the horizontal and vertical angular position of the target object, providing an accurate basis for subsequent target positioning. Radar cross-sectional area (RCS) signature calculation is based on signal reflection intensity. Targets of different materials and shapes reflect different millimeter-wave signals. By analyzing the amplitude and phase information of the reflected signal, the target's RCS signature is estimated, allowing a preliminary assessment of its size and type. Furthermore, to extract target shape signatures, the module employs time-frequency analysis and spatial spectrum estimation algorithms based on millimeter-wave signals. Time-frequency analysis can decompose signals in two dimensions: time and frequency, capturing subtle characteristic changes of target objects in different motion states. The spatial spectrum estimation algorithm reconstructs the shape outline of the target in the spatial spectrum domain based on the differences in the reflection characteristics of different target objects to millimeter-wave signals, thereby achieving accurate extraction of the target shape, for example, distinguishing the outlines of pedestrians and vehicles.

[0045] (4) Roadblock Classification Module

[0046] The roadblock classification module is a core component of the system. It builds a classification model based on a convolutional neural network (CNN). The module is pre-trained using a large dataset containing sample features of different types of roadblocks.

[0047] The dataset covers millimeter-wave feature data for a variety of typical roadblock targets, including fixed obstacles (such as road guardrails and isolation piers), moving vehicles (including cars, trucks, buses, and other vehicles), pedestrians (pedestrian models in different postures and wearing different clothing), and traffic facilities (traffic signs, signal lights, etc.). During training, multiple convolutional layers perform convolution operations on the input feature data to extract deep feature representations. Pooling layers are used to reduce feature dimensionality, reduce computational complexity, and enhance the model's translational invariance to features. Fully connected layers integrate the extracted features and ultimately output the category to which the target object belongs. After training on a large amount of sample data, the CNN model can automatically learn the characteristic patterns of different types of roadblocks and achieve high-precision classification and recognition of roadblock types, with a classification accuracy rate exceeding 95%.

[0048] (5) Path Planning Module

[0049] The path planning module, integrated into the autonomous driving control software of the smart car and connected to the roadblock classification module, replans the smart car's route in real time based on the type and location of the roadblock, as well as the smart car's current position (determined by its onboard GPS and inertial navigation system) and direction of travel.

[0050] Path planning utilizes the advanced A-star algorithm, which comprehensively considers factors such as the location, size, and shape of roadblocks, as well as the vehicle's kinematic constraints (such as minimum turning radius and maximum speed) and the driving environment (such as road width and curvature). Using shortest path length, highest driving safety, optimal path smoothness, and lowest energy consumption as evaluation criteria, the system uses a heuristic search strategy to identify an optimal path within the drivable area surrounding the vehicle. During the planning process, roadblocks are treated as obstruction points or areas, ensuring that the planned path maintains a safe distance from obstacles, providing reliable guidance for the vehicle.

[0051] (6) Control module

[0052] The control module is a key component in the system's interaction with the smart car's hardware system and is closely connected to the path planning module. Its function is to generate corresponding control instructions based on the planned driving path, driving the smart car's steering system, powertrain, and other systems, enabling the smart car to accurately follow the planned path.

[0053] The system utilizes a Model Predictive Control (MPC) algorithm, based on the vehicle's dynamics model, to predict the vehicle's future motion over time in real time. By continuously optimizing control commands, such as adjusting steering angle and vehicle speed, the vehicle's actual trajectory is kept as close to the planned path as possible. When avoiding obstacles, the MPC algorithm fully considers the vehicle's dynamic characteristics, ensuring stable driving performance during emergency avoidance operations. This prevents loss of control due to sudden steering or speed changes, effectively improving driving comfort and safety.

[0054] (7) Data fusion module

[0055] As an extension of the system, the data fusion module is connected to the millimeter-wave radar sensor, the smart car's camera, the ultrasonic sensor, and the on-board communication system, playing a key role in integrating multi-source information.

[0056] The module first synchronizes and transforms the data from different sensors to ensure temporal and spatial consistency. For example, the target position information detected by the millimeter-wave radar is converted to the same coordinate system as the camera image to facilitate data fusion. During the fusion process, a combination of algorithms, including Kalman filter fusion and particle filter fusion, is employed. The Kalman filter is suitable for linear systems, enabling optimal estimation of the target state and smoothing of filtered data. The particle filter excels in complex systems with nonlinear and non-Gaussian noise. It uses a large number of particle samples to describe the probability density function of the target state, improving estimation accuracy in complex scenarios. Data fusion leverages the strengths of each sensor, such as millimeter-wave radar in long-range detection and speed measurement, and cameras in target visual recognition. This allows for a more comprehensive and accurate perception of the surrounding environment, improving the accuracy and reliability of target detection and obstacle identification, and providing a more robust information foundation for path planning and control.

[0057] 2. Further explanation of the above features (1) Millimeter wave radar sensor installation and calibration

[0058] A millimeter-wave radar sensor is installed inside the left and right bumpers of the smart car. It is mounted approximately 30 centimeters from the ground and tilted 15 degrees forward, ensuring coverage within a 120-degree radius directly in front of the vehicle and within 200 meters of the vehicle. A millimeter-wave radar sensor is also installed on the side of the door, at the front and rear edges, approximately 40 centimeters from the ground and mounted horizontally to the side. It detects targets within a 90-degree radius to the side of the vehicle and within 50 meters of the vehicle.

[0059] After installation, the millimeter-wave radar sensor undergoes precise calibration. Using specialized calibration equipment, in an open, interference-free environment, the sensor's parameters, such as transmit power, receive sensitivity, and beam pointing, are adjusted to ensure optimal performance. Furthermore, multiple measurements are performed against a standard target of known position and size (such as a calibration reflector) to correct for measurement errors and improve the accuracy of distance, velocity, and angle measurements.

[0060] (2) Signal processing module algorithm optimization

[0061] The adaptive filtering algorithm uses a recursive implementation based on the Minimum Mean Square Error (MMSE) criterion. The module monitors the power spectral density of ambient noise in real time and automatically adjusts the filter weights based on the set convergence coefficient. In strong interference environments, the filter converges quickly, effectively suppressing noise interference and improving the signal-to-noise ratio by over 20dB.

[0062] The analog-to-digital converter utilizes a high-speed 12-bit ADC with a sampling frequency of 2.5MHz, ensuring signal bandwidth requirements while providing high-precision digital signal output. To further improve signal quality, an automatic gain control (AGC) circuit is added before the analog-to-digital conversion. This circuit automatically adjusts the amplification factor based on the signal amplitude, ensuring that the signal amplitude entering the ADC remains stable within the optimal quantization range and minimizing signal distortion.

[0063] (3) Feature extraction module deep feature mining

[0064] To extract target shape features, in addition to basic time-frequency analysis and spatial spectrum estimation algorithms, a wavelet transform algorithm is introduced to perform multi-scale analysis of millimeter-wave signals. Wavelet transforms decompose signals into different scales for processing, capturing detailed shape features of the target at different scales, such as edge features and local concavity and convexity of the target's contour. Combined with the spatial spectrum estimation algorithm, this further improves the accuracy and resolution of target shape feature extraction, enabling the system to more accurately identify different types of targets, such as different types of moving vehicles or pedestrians with different postures.

[0065] The extraction of RCS features considers the impact of the target's posture changes on the RCS value. By building a 3D model of the target object, its RCS feature values ​​are simulated and calculated at different posture angles. This value is stored in a database as prior knowledge. During the actual detection process, the RCS features are corrected based on the extracted RCS feature values ​​and the target posture estimation results to improve the accuracy of target type judgment.

[0066] (4) Roadblock Classification Module Model Optimization and Training

[0067] The convolutional neural network model is an optimization and improvement based on the traditional CNN architecture. In the convolutional layer, depthwise separable convolution is used instead of standard convolution, reducing the number of model parameters and computational complexity, improving model efficiency, and maintaining efficient feature extraction. In the pooling layer, a combination of max pooling and average pooling is used to preserve significant information in the target features while smoothing the feature data, enhancing the model's robustness to the target features.

[0068] During training, data augmentation techniques were used to expand the sample data. The original sample data was subjected to random rotation, translation, scaling, and noise addition to simulate changes in target features under various road and environmental conditions. The resulting augmented sample data volume reached five times the original volume. Furthermore, leveraging the hierarchical structure of the data labels, a hierarchical softmax loss function was introduced into the loss function, enabling the model to better learn the hierarchical relationships between target categories, improving classification accuracy and generalization. After 100 epochs of training, the model achieved classification accuracy exceeding 96% on the validation set, with a loss value below 0.1, meeting the requirements of practical applications.

[0069] (5) Path planning module environment modeling and optimization

[0070] Before planning a path, the module first constructs an environmental model of the smart car's surroundings based on roadblock information detected by the millimeter-wave radar sensor and the vehicle's position information provided by the onboard GPS and inertial navigation system. This environmental model uses a grid map representation, dividing the driving area into small grid cells. Each cell is marked according to whether it is occupied by a roadblock or drivable. The grid map has a resolution of 0.1 meter by 0.1 meter, accurately reflecting the details of the surrounding environment.

[0071] To improve the efficiency and quality of path planning, the A* algorithm has been optimized and improved. In addition to considering the Euclidean distance of the path, the heuristic function also incorporates factors such as changes in path curvature, slope, and distance from roadblocks as weight coefficients. By continuously adjusting the weight coefficients, the algorithm is more inclined to select smooth, safe paths that comply with vehicle dynamics constraints when searching for paths. At the same time, the concept of dynamic programming is adopted to record and update path nodes that have been searched, avoiding repeated searches and improving the algorithm's operating efficiency. In complex road conditions, such as multi-lane intersections and road construction scenarios, the optimized A* algorithm can plan an optimal driving path that meets the requirements within 0.2 seconds, providing strong support for real-time obstacle avoidance of smart cars.

[0072] (6) Control module precise instruction generation and execution

[0073] The model predictive control algorithm establishes a precise dynamic model of the intelligent vehicle, including a description of the dynamic characteristics of the vehicle's powertrain, steering system, and braking system. By collecting real-time vehicle status information (such as speed, steering angle, acceleration, etc.), combined with the target path provided by the path planning module, the vehicle's trajectory within the next two seconds is predicted. Based on the predicted results, a quadratic programming algorithm is used to optimize control instructions, ensuring that the vehicle's actual trajectory closely matches the target path while satisfying vehicle stability and safety constraints.

[0074] When generating control commands, the nonlinear characteristics of the vehicle's steering system and the delay characteristics of its powertrain are comprehensively considered. By combining feedforward control with feedback compensation control, the accuracy and real-time performance of control commands are improved. Feedforward control generates steering angle and speed commands in advance based on the curvature and slope of the target path. Feedback compensation control adjusts control commands in real time based on the deviation between the vehicle's actual response and the expected response, eliminating the impact of system delays and nonlinearities. Control commands are transmitted via the vehicle's CAN bus to actuators such as the steering system, powertrain, and braking system. Upon receiving the commands, the actuators rapidly respond and adjust the vehicle's driving state, ensuring that the smart car smoothly and accurately follows the planned path. In actual tests, the vehicle's lateral deviation during obstacle avoidance was kept within 0.3 meters, and the longitudinal speed control error was within 2%, effectively improving driving comfort and safety.

[0075] 3. System Operation Process

[0076] (1) System initialization

[0077] 1. When a smart car starts, the millimeter-wave-based smart car roadblock recognition system begins a self-test. The millimeter-wave radar sensor checks the operating status of each transmitting and receiving antenna to ensure proper signal transmission and reception, and performs initial beam calibration. The signal processing module initializes filter parameters, resets the analog-to-digital converter, and prepares to receive and process radar signals. The feature extraction module loads pre-stored feature extraction algorithm parameters and initializes the cache required for time-frequency analysis and spatial spectrum estimation algorithms. The roadblock classification module loads the trained convolutional neural network model weight parameters and prepares for target classification. The path planning module initializes map data and the open and closed lists required by the A* algorithm. The control module initializes the vehicle dynamics model parameters used in the model predictive control algorithm and establishes communication connections with the vehicle's actuators. The data fusion module checks the connection status with other sensors (cameras, ultrasonic sensors, etc.) and the on-board communication system, and initializes the data reception and fusion processing buffers.

[0078] 2. After the system self-test completes, each module enters standby mode, awaiting driving instructions from the vehicle. During this time, the millimeter-wave radar sensor operates in low-power listening mode, monitoring the surrounding millimeter-wave signal in real time. If it detects an approaching target or other abnormal signal, it immediately triggers the system's rapid response mechanism, waking up other modules and putting them into operation.

[0079] (2) Normal operation stage

[0080] 1. Once a vehicle begins moving, the millimeter-wave radar sensor transmits millimeter-wave signals in real time toward the front and sides of the vehicle at a preset operating frequency and transmit power. When the signal encounters a target object (such as a vehicle ahead, pedestrians, or road guardrails), it is reflected and captured by the radar sensor's receiving antenna.

[0081] 2. After receiving the reflected signal, the signal processing module first uses an adaptive filtering algorithm to filter the signal to remove environmental noise. It then amplifies the filtered signal and adjusts the signal amplitude to an appropriate range. An analog-to-digital converter then converts the analog signal into a digital signal, extracting the original feature information of the target object. This generates a digital signal feature sequence and sends it to the feature extraction module.

[0082] 3. After receiving the digital signal feature sequence, the feature extraction module uses a time-frequency analysis algorithm to perform a time-frequency transformation on the signal to extract the target's velocity characteristics. Simultaneously, an angle-of-arrival and angle-of-departure estimation algorithm, combined with information from multiple receiving channels of the MIMO radar array, determines the target's angular position characteristics. By analyzing the signal's amplitude and phase information, the target's distance and RCS characteristics are calculated. Furthermore, wavelet transforms and spatial spectrum estimation algorithms are used to deeply mine and extract the target's shape features, generating a target feature vector containing multiple features. This vector is then passed to the obstacle classification module.

[0083] 4. After receiving the target feature vector, the roadblock classification module inputs it into a trained convolutional neural network model. Through layer-by-layer processing using convolutional, pooling, and fully connected layers, the model automatically extracts deep information from the target features and matches and compares it with the characteristic patterns of various roadblocks learned during training. Ultimately, it outputs the roadblock type to which the target object belongs—for example, a stationary car (fixed obstacle) 50 meters ahead or a pedestrian crossing the road to the side—and sends the classification result to the path planning module.

[0084] 5. The path planning module uses the obstacle type and location information provided by the obstacle classification module, combined with the smart car's current driving status (real-time vehicle position and speed information obtained through the onboard GPS and inertial navigation system), to replan the smart car's driving path in real time within a pre-built environmental model using an optimized A* algorithm. During the planning process, factors such as the safe distance between the obstacle and the vehicle, road traffic regulations, and the vehicle's kinematic constraints are comprehensively considered to ensure that the planned path is both safe and feasible. The generated planned path is transmitted to the control module as a series of target points. The obstacle information and planned path are simultaneously updated in real time on the onboard display screen, providing intuitive driving assistance information to the driver.

[0085] 6. After receiving the planned path, the control module uses a model predictive control algorithm, combined with the smart car's dynamic model, to predict the vehicle's motion state over the next period of time. Based on the prediction results, precise control instructions are generated in real time, including steering angle and speed adjustment instructions. The control instructions are sent to actuators such as the steering system and power system via the vehicle's CAN bus. The steering system precisely adjusts the vehicle's front wheel steering angle based on the received steering angle instruction; the power system controls the engine's output power or the motor's torque based on the speed adjustment instruction, allowing the smart car to smoothly and accurately follow the planned path and effectively avoid roadblocks. Throughout the driving process, the control module continuously receives feedback on the vehicle's actual driving state, compares it with the planned path in real time, and promptly adjusts the control instructions based on any deviations to ensure that the vehicle always follows the planned path.

[0086] 7. Throughout its operation, the data fusion module continuously receives data from multiple sources, including millimeter-wave radar sensors, cameras, ultrasonic sensors, and onboard communication systems. After time synchronization and coordinate conversion, this data is fused using Kalman filter fusion and particle filter fusion algorithms. The fused data more comprehensively and accurately reflects the environmental information surrounding the vehicle, further improving the reliability of target detection and roadblock identification. For example, when the millimeter-wave radar detects a target ahead but cannot accurately determine its type, the data fusion module can combine the image information captured by the camera and use an image recognition algorithm to assist in determining the target type, thereby improving the accuracy of roadblock classification. Furthermore, the fused data provides richer information support for the path planning and control modules, helping to optimize path planning and control strategies.

[0087] (3) Emergency handling

[0088] 1. When the obstacle classification module determines that the obstacle's hazard level is high, such as when a large, scattered obstacle suddenly appears on the road ahead and the vehicle is traveling at a high speed, potentially preventing a collision through conventional braking or maneuvering. The system immediately triggers the emergency braking sequence. The control module quickly generates a full-strength braking command and sends it to the braking system via the vehicle's CAN bus. Upon receiving the command, the braking system immediately activates the vehicle's emergency braking function, applying maximum braking force to rapidly decelerate the vehicle. Simultaneously, the vehicle's hazard light automatically illuminates, alerting vehicles behind to evade the vehicle.

[0089] 2. During full braking, the system continuously monitors changes in the distance between the vehicle and the obstacle, as well as changes in vehicle speed. If the distance is too close and braking distance is insufficient to avoid a collision, the path planning module will attempt to find an alternative evasive path, such as an emergency lane change to a safe lateral area. Based on the new planned path, the control module rapidly adjusts the steering angle and performs an emergency lane change while applying full braking, minimizing or avoiding a collision with the obstacle.

[0090] 3. Once the vehicle successfully avoids an obstacle or comes to a complete stop, the system automatically releases the emergency brake and returns to normal driving mode. The path planning module replans a safe driving path, and the control module generates control instructions based on the new path, allowing the vehicle to continue driving smoothly. Simultaneously, the system records relevant data about the emergency (such as obstacle type, location, and vehicle driving status) in the vehicle's black box, providing data support for subsequent safety analysis and system optimization.

[0091] IV. System Performance Test and Results

[0092] (1) Target detection and classification performance test

[0093] 1. Test Method: A variety of typical road scenarios were set up at the intelligent vehicle technology test site, including urban roads, highways, and rural roads. Within these scenarios, different types of target objects were placed, such as fixed obstacles (guardrails, traffic signs, etc.), moving vehicles (sedans, SUVs, trucks, etc. traveling at different speeds), pedestrians (dummies simulating different walking postures and speeds), and traffic facilities (traffic lights, electronic signs, etc.). An intelligent vehicle equipped with this roadblock recognition system was tested multiple times within the test site, following a set route and speed. During driving, the target data detected by the millimeter-wave radar sensor and the classification results of the roadblock classification module were recorded. The classification results were compared and analyzed with the manually labeled real-world target types, and the classification accuracy and recall rate were calculated.

[0094] 2. Test Results: After testing 1,000 target object samples, the system achieved 97% accuracy and 95% recall for fixed obstacles; 96% accuracy and 93% recall for moving vehicles; 94% accuracy and 90% recall for pedestrians; and 98% accuracy and 96% recall for traffic facilities. The overall classification accuracy reached 95.5% and the recall rate was 93.8%. These test results demonstrate that the system can accurately identify various typical roadblocks, providing a reliable basis for autonomous decision-making in intelligent vehicles.

[0095] (2) Path planning and control performance test

[0096] 1. Test Method: Also at the smart car technology test site, complex traffic congestion scenarios and emergency obstacle avoidance scenarios were simulated. In the traffic congestion scenario, multiple test vehicles randomly stopped or drove at low speeds to form a narrow, passable passage. In the emergency obstacle avoidance scenario, pedestrians suddenly appeared crossing the road or vehicles drove out from behind the adjacent vehicles. The smart car was driven through these scenarios at different initial speeds (such as 30km / h, 50km / h, and 80km / h), and the system-planned driving path, the vehicle's actual driving trajectory, and the success or failure of obstacle avoidance were recorded. At the same time, indicators such as the vehicle's lateral deviation, longitudinal speed control error, and obstacle avoidance reaction time during the obstacle avoidance process were measured.

[0097] 2. Test Results: In traffic congestion scenarios, the system successfully planned a path through narrow passages. The average lateral deviation between the vehicle's actual trajectory and the planned path was 0.15 meters, the longitudinal speed control error was within 3%, and the obstacle avoidance success rate reached 98%. In emergency obstacle avoidance scenarios, at an initial speed of 30 km / h, the obstacle avoidance reaction time was 0.18 seconds, the lateral deviation was 0.2 meters, and the obstacle avoidance success rate was 100%. At an initial speed of 50 km / h, the obstacle avoidance reaction time was 0.22 seconds, the lateral deviation was 0.25 meters, and the obstacle avoidance success rate was 96%. At an initial speed of 80 km / h, the obstacle avoidance reaction time was 0.25 seconds, the lateral deviation was 0.3 meters, and the obstacle avoidance success rate was 92%. These test results demonstrate that the system has excellent path planning and control performance, effectively guiding smart cars to avoid obstacles in complex road conditions, ensuring driving safety.

[0098] (3) Data fusion performance test

[0099] 1. Test Method: Multiple millimeter-wave radar reflective targets, visual marker targets, and ultrasonic reflective targets were placed simultaneously in a specific area of ​​the test field. A smart car was driven at a specific speed within this area, and the data fusion module's results from the millimeter-wave radar sensor, camera, and ultrasonic sensor were recorded. By comparing the individual sensor detection results with the fused results, the data fusion module's performance improvements in target detection accuracy, false alarm rate, and missed detection rate were evaluated.

[0100] 2. Test Results: The target detection accuracy of a single millimeter-wave radar sensor was 90%, with an 8% false alarm rate and a 10% missed detection rate. The target detection accuracy of a single camera was 85%, with a 12% false alarm rate and a 15% missed detection rate. The target detection accuracy of a single ultrasonic sensor was 80%, with a 15% false alarm rate and a 20% missed detection rate. After processing by the data fusion module, target detection accuracy increased to 96%, with a 3% false alarm rate and a 5% missed detection rate. This demonstrates that the data fusion module effectively integrates multi-source sensor data, significantly improving the accuracy and reliability of target detection and enhancing the system's ability to perceive complex environments.

[0101] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A millimeter wave-based intelligent automobile roadblock recognition system, characterized in that: include: Millimeter-wave radar sensors, installed on the front and sides of smart cars, are used to transmit millimeter-wave signals and receive reflected signals; The signal processing module is connected to the millimeter wave radar sensor and is used to pre-process the reflected signal, including filtering, amplification and analog-to-digital conversion operations, to extract the original feature information of the target object; The feature extraction module is connected to the signal processing module and is used to extract various features of the target object from the pre-processed reflection signal, including distance, speed, angle, radar cross-section area and target shape features; The roadblock classification module, connected to the feature extraction module, is built based on a machine learning algorithm. A classification model is pre-trained using a dataset containing sample features of different types of roadblocks. The module is used to classify and identify target objects based on the extracted target features, determining whether they are roadblocks and the specific type of roadblock. Roadblock types include fixed obstacles, moving vehicles, pedestrians, and traffic facilities. The path planning module is connected to the roadblock classification module. It replans the smart car's driving path in real time based on the type and location of the roadblock and the current position and driving direction of the smart car to avoid the roadblock and ensure driving safety. The control module is connected to the path planning module and is used to generate corresponding control instructions according to the planned driving path, control the driving direction and speed of the smart car, and make the smart car drive according to the planned path.

2. The intelligent automobile roadblock recognition system based on millimeter waves according to claim 1 is characterized in that: The millimeter-wave radar sensor includes multiple transmitting antennas and multiple receiving antennas to form a multi-input multi-output radar array, which can improve the resolution and accuracy of target detection and achieve simultaneous detection and tracking of multiple targets; The signal processing module adopts an adaptive filtering algorithm, which can automatically adjust the filter parameters according to the changes in environmental noise, effectively suppress noise interference and improve signal quality; The method for extracting target shape features by the feature extraction module includes a time-frequency analysis and spatial spectrum estimation algorithm based on millimeter wave signals, which can accurately extract the shape contour features of the target according to the differences in the reflection characteristics of different target objects to millimeter wave signals; The roadblock classification module uses a convolutional neural network to build a classification model. The convolutional neural network classification model includes multiple convolutional layers, pooling layers and fully connected layers. By training on a large amount of sample data labeled with different types of roadblock features, it can automatically learn the feature representation of the target object and achieve high-precision classification and recognition of roadblock types.

3. The intelligent automobile roadblock recognition system based on millimeter waves according to claim 1 is characterized in that: It also includes a data fusion module, which is connected to the millimeter-wave radar sensor, the smart car's camera, the ultrasonic sensor and the on-board communication system respectively, and is used to fuse data from different sensors and communication equipment, including Kalman filter fusion, particle filter fusion and other fusion algorithms to improve the accuracy and reliability of target detection and roadblock identification.

4. The millimeter wave-based intelligent automobile roadblock recognition system according to claim 1 is characterized in that: The control module adopts a model predictive control algorithm to predict the future state of the smart car in real time based on the planned driving path and the dynamic model of the smart car, and generates corresponding control instructions, so that the smart car can avoid road obstacles while maintaining stable driving performance and improving driving comfort.

5. A millimeter wave-based intelligent vehicle roadblock recognition method, characterized in that: The following steps are involved: Transmitting millimeter wave signals and receiving reflected signals through a millimeter wave radar sensor; Preprocess the reflected signal to extract the original feature information of the target object; Extracting multiple features of the target object from the preprocessed reflection signal; Classify and identify the target object based on the extracted target features to determine whether it is a roadblock and the specific type of roadblock; Replan the smart car's route in real time based on the type and location of the roadblock and the smart car's current location and direction of travel; Generate corresponding control instructions based on the planned driving path to control the driving direction and speed of the smart car.

6. The millimeter wave-based intelligent automobile roadblock recognition method according to claim 5, characterized in that: Before preprocessing the reflected signal, the process also includes time synchronization and coordinate conversion steps for the data collected by the millimeter-wave radar sensor to ensure the temporal and spatial consistency of data collected by different sensors and improve the accuracy of data fusion and target recognition.

7. The millimeter wave-based intelligent vehicle roadblock recognition method according to claim 5, characterized in that: The extraction of multiple features of the target object includes calculating the speed of the target object based on the Doppler effect of the millimeter wave signal, estimating the angular position of the target object based on the signal arrival angle and departure angle, and calculating the distance and RCS characteristics of the target object through the amplitude and phase information of the signal.

8. The millimeter wave-based intelligent automobile roadblock recognition method according to claim 5, characterized in that: After classifying and identifying the target object, the system also includes calculating the danger level of the roadblock based on the classification results, and adjusting the alarm level and control strategy according to the danger level. When the danger level of the roadblock is high, the emergency braking system of the smart car is triggered in time or an audible and visual alarm prompt is issued to ensure driving safety.

9. The method for intelligent automobile roadblock recognition based on millimeter waves according to claim 5, characterized in that: The corresponding control instructions are generated according to the planned driving path, including adjusting the steering angle and speed of the smart car according to the curvature and slope of the path, so that the smart car can maintain a reasonable driving speed and stable handling performance while avoiding road obstacles, thereby improving driving safety and comfort.

10. A smart car, characterized in that: The intelligent automobile roadblock recognition system based on millimeter waves as described in any one of claims 1 to 4 is installed.