Millimeter wave radar obstacle detection method and system for intelligent traffic

Abstract

The invention discloses a millimeter-wave radar obstacle detection method and system for intelligent traffic, and relates to the technical field of intelligent traffic, and the method comprises the steps: S1, obtaining an original radar signal collected by a millimeter-wave radar, S2, carrying out the preprocessing of the original radar signal based on a multi-path reflection model, so as to filter out a false target caused by the reflection of a high building, and carrying out the detection of an obstacle through a millimeter-wave radar. S3, carrying out obstacle identification and classification on the preprocessed radar signals by using a deep learning algorithm, and S4, generating obstacle information according to identification and classification results; according to the millimeter wave radar obstacle detection method and system for intelligent traffic, the problem of false target interference caused by high-rise reflection in a complex urban environment in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation technology, and in particular to a millimeter wave radar obstacle detection method and system for intelligent transportation. Background Art

[0002] In the current intelligent transportation field, millimeter wave radar, as a key sensor technology, is widely used in vehicle obstacle detection and environmental perception due to its all-weather and high-precision characteristics. Millimeter wave radar detects targets in the surrounding environment by emitting and receiving high-frequency radio waves, and can provide target distance, speed and direction information, thereby realizing real-time obstacle detection function.

[0003] However, in complex urban environments, the performance of millimeter-wave radars will be significantly interfered with. These interferences mainly come from high-rise buildings, large glass curtain walls and other highly reflective surfaces in urban environments. When millimeter-wave radar signals encounter these reflective surfaces, multipath propagation may occur, causing the echo signals received by the radar to contain false target information. These false targets will interfere with the judgment of the obstacle detection system, thereby affecting the accuracy and reliability of the detection results. In the prior art, some solutions attempt to filter out false target interference through signal processing algorithms, but these methods usually require a lot of computing resources, and it is still difficult to completely eliminate false targets in dynamically changing complex urban environments. In addition, the problem of multipath interference caused by reflections from high-rise buildings has not yet been fundamentally solved, especially in high-density building areas, the number and intensity of false targets will increase significantly, causing the performance of the obstacle detection system to drop significantly. Summary of the invention

[0004] The purpose of the present invention is to provide a millimeter wave radar obstacle detection method and system for intelligent transportation, so as to solve the problem of false target interference caused by reflection from high buildings in complex urban environments in the prior art.

[0005] To achieve the above object, the present invention provides the following technical solution: a millimeter wave radar obstacle detection method for intelligent transportation, the method comprising:

[0006] S1. Obtaining the original radar signal collected by the millimeter wave radar;

[0007] S2. Based on the multipath reflection model, the original radar signal is preprocessed to filter out false targets caused by reflections from high-rise buildings, including decomposing the signal into long-term trends, seasonal components, and noise residuals, extracting the trend component as the true target signal, removing the high-frequency seasonal component, and outputting the signal to reduce false targets caused by reflection interference. The specific formula for decomposing the signal is:

[0008] y[n]=T[n]+S[n]+R[n];

[0009] Where y[n] represents the original signal collected by the millimeter-wave radar, T[n] represents the long-term trend component in the signal, S[n] represents the seasonal component in the signal, R[n] represents the noise residual in the signal, and n represents the sampling index of the millimeter-wave radar collected signal;

[0010] The specific formula of the output signal is: z[n]=y[n]-S[n];

[0011] Wherein, z[n] represents the effective signal after preprocessing, y[n] represents the original signal collected by the millimeter-wave radar, S[n] represents the seasonal component in the signal, and n represents the sampling index of the millimeter-wave radar collected signal;

[0012] S3, using deep learning algorithms to identify and classify obstacles on preprocessed radar signals;

[0013] S4. Generate obstacle information based on the recognition and classification results.

[0014] Preferably, the S1 comprises:

[0015] Get the original radar signal and design a bandpass filter to extract the target frequency band in the signal. The specific formula is: g(t) = x(t) × h(t);

[0016] Among them, g(t) represents the target signal obtained after filtering, x(t) represents the original signal collected by the millimeter-wave radar, h(t) represents the characteristic function of the bandpass filter, and t represents time.

[0017] Preferably, S3 includes:

[0018] Define the classification model. For each radar signal sample, use the trained logistic regression model to predict the probability of it belonging to a certain category. The specific formula is:

[0019]

[0020] Where p represents the probability value of the millimeter-wave radar signal being classified as a certain type of obstacle, w represents the weight parameter vector of the model, k represents the preprocessed feature extracted from the millimeter-wave radar signal, b represents the bias term, and e represents the cardinality of the natural exponential function;

[0021] Set the classification threshold to complete obstacle classification.

[0022] Preferably, S4 includes:

[0023] Define the priority scoring index of obstacle information and calculate the priority score of each obstacle. The specific formula is: S = m1×d+m2×u+m3×v;

[0024] Among them, S represents the comprehensive score of the obstacle, m1 represents the distance weight, m2 represents the size weight, m3 represents the speed weight, d represents the distance between the obstacle and the millimeter-wave radar, u represents the size of the obstacle, and v represents the speed of the obstacle;

[0025] The obstacles are sorted according to the scores and corresponding handling strategies are generated.

[0026] Preferably, the ordering of obstacles according to scores and generating corresponding processing strategies include setting a threshold Q, triggering an alarm or adjusting the vehicle path when S>Q, and only recording information or not processing when S≤Q.

[0027] Preferably, the setting of the classification threshold to complete obstacle classification includes: when p>0.5, judging it as a vehicle, and when p≤0.5, judging it as a pedestrian.

[0028] Preferably, S3 further includes:

[0029] The preprocessed signal features are processed by dimensionality reduction algorithm, the specific formula is: r′=Y×r;

[0030] Among them, r′ represents the eigenvector after dimensionality reduction, Y represents the dimensionality reduction matrix, and r represents the original eigenvector.

[0031] Preferably, the operating frequency range of the millimeter wave radar is 24 GHz to 77 GHz.

[0032] Preferably, the S4 further includes:

[0033] When generating obstacle information, it includes the type, position and motion state of the obstacle. The motion state is determined by calculating the acceleration of the obstacle. The specific formula is: a = (l t+1 -l t ) / Δt;

[0034] Where a represents the acceleration of the obstacle, l t+1 represents the speed of the obstacle at time t+1, l t represents the speed of the obstacle at time t, t represents time, and Δt represents the time interval.

[0035] A millimeter wave radar obstacle detection system for intelligent transportation, used to implement the steps of the millimeter wave radar obstacle detection method for intelligent transportation, the system comprising:

[0036] A radar signal acquisition module is used to obtain the original radar signal collected by the millimeter-wave radar;

[0037] A signal preprocessing module connected to the radar signal acquisition module, used to preprocess the original radar signal based on a multipath reflection model to filter out false targets caused by reflections from high-rise buildings;

[0038] The obstacle recognition and classification module connected to the signal preprocessing module is used to use the deep learning algorithm to perform obstacle recognition and classification on the preprocessed radar signal;

[0039] The information generation module connected to the obstacle recognition and classification module is used to generate obstacle information based on the recognition and classification results.

[0040] It can be seen from the above technical solution that the present invention has the following beneficial effects:

[0041] The millimeter-wave radar obstacle detection method and system for intelligent transportation obtains the original radar signal collected by the millimeter-wave radar, pre-processes the original radar signal based on the multipath reflection model to filter out the false targets caused by the reflection of high-rise buildings, uses the deep learning algorithm to identify and classify the pre-processed radar signal, and generates obstacle information according to the identification and classification results. It can significantly reduce the interference of false targets in the radar signal, thereby improving the accuracy and reliability of obstacle detection, and can adapt to the complex and changeable urban environment in real time, especially in the high-density building area. It can still maintain efficient and stable detection capabilities, effectively reduce the calculation complexity, improve the system operation efficiency, and can accurately identify and effectively filter out the periodic and high-intensity false targets caused by the reflection of high-rise buildings. It can effectively handle fast-moving targets in the urban environment, improve the recognition and classification capabilities of dynamic targets, and solve the problem of false target interference caused by the reflection of high-rise buildings in the existing technology in complex urban environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a flow chart of the method of the present invention;

[0043] Figure 2 It is a schematic diagram of the connection of the system modules of the present invention. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] like Figure 1 As shown, the present invention provides a technical solution: a millimeter wave radar obstacle detection method for intelligent transportation, the method comprising:

[0046] S1. Obtaining the original radar signal collected by the millimeter wave radar;

[0047] S2. Based on the multipath reflection model, the original radar signal is preprocessed to filter out false targets caused by reflections from high-rise buildings, including decomposing the signal into long-term trends, seasonal components, and noise residuals, extracting the trend component as the true target signal, removing the high-frequency seasonal component, and outputting the signal to reduce false targets caused by reflection interference. The specific formula for decomposing the signal is:

[0048] y[n]=T[n]+S[n]+R[n];

[0049] Where y[n] represents the original signal collected by the millimeter-wave radar, T[n] represents the long-term trend component in the signal, S[n] represents the seasonal component in the signal, R[n] represents the noise residual in the signal, and n represents the sampling index of the millimeter-wave radar collected signal;

[0050] The specific formula of the output signal is: z[n]=y[n]-S[n];

[0051] Wherein, z[n] represents the effective signal after preprocessing, y[n] represents the original signal collected by the millimeter-wave radar, S[n] represents the seasonal component in the signal, and n represents the sampling index of the millimeter-wave radar collected signal;

[0052] S3, using deep learning algorithms to identify and classify obstacles on preprocessed radar signals;

[0053] S4. Generate obstacle information based on the recognition and classification results.

[0054] In the above implementation, the method first obtains the original signal collected by the millimeter wave radar, and pre-processes the signal based on the multipath reflection model to reduce the interference of false targets caused by environmental reflections such as high-rise buildings. In the signal preprocessing, the signal decomposition method is used to divide the original signal into a long-term trend component, a seasonal component and a noise residual. The long-term trend component represents the real target signal, and the seasonal component and the noise residual are removed to extract the effective signal. Subsequently, the processed signal is identified and classified by a deep learning algorithm. The deep learning algorithm can effectively distinguish different types of obstacles and improve the classification accuracy through a large amount of training data. Finally, the corresponding obstacle information is generated according to the results of identification and classification for subsequent processing and decision-making by the intelligent transportation system. The method significantly reduces the false targets caused by environmental reflections through the multipath reflection model and signal decomposition technology, and improves the accuracy and reliability of signal processing. The use of a deep learning algorithm can realize efficient classification of different obstacles, and improve the recognition ability and real-time performance of the intelligent transportation system in complex environments. In addition, the generated obstacle information is highly accurate, providing reliable data support for subsequent intelligent transportation control, thereby improving the safety, stability and applicability of the intelligent transportation system.

[0055] S1 includes obtaining the original radar signal and designing a bandpass filter to extract the target frequency band in the signal. The specific formula is: g(t) = x(t) × h(t);

[0056] Among them, g(t) represents the target signal obtained after filtering, x(t) represents the original signal collected by the millimeter-wave radar, h(t) represents the characteristic function of the bandpass filter, and t represents time.

[0057] In the above implementation, in order to highlight the characteristics of the target frequency band signal and reduce the interference of noise and irrelevant frequency bands, the collected original signal x(t) is filtered by designing a bandpass filter h(t) for the multi-frequency components contained in the millimeter-wave radar signal. The characteristic function h(t) of the bandpass filter is optimized according to the center frequency and bandwidth of the target frequency band to ensure that the target frequency band characteristics in the signal can be effectively extracted while suppressing the interference components in other frequency ranges. The filtered signal g(t) only retains the frequency components of the target frequency band, which greatly improves the quality and effectiveness of the signal and lays the foundation for subsequent signal processing steps (such as decomposition, classification and identification). By designing a bandpass filter in the signal preprocessing stage, the target frequency band characteristics in the millimeter-wave radar signal can be effectively extracted, and the influence of environmental noise and irrelevant frequency band interference can be reduced. Compared with the traditional method, this method improves the signal-to-noise ratio and spectral separation of the target signal, making the subsequent obstacle identification and classification more accurate. In addition, the flexible design of the bandpass filter can adjust the parameters according to different application requirements, so as to adapt to a variety of complex environments, and greatly improve the reliability and versatility of the millimeter-wave radar system.

[0058] S3 includes a model that defines classification. For each radar signal sample, the trained logistic regression model is used to predict the probability that it belongs to a certain category. The specific formula is:

[0059]

[0060] Where p represents the probability value of the millimeter-wave radar signal being classified as a certain type of obstacle, w represents the weight parameter vector of the model, k represents the preprocessed feature extracted from the millimeter-wave radar signal, b represents the bias term, and e represents the cardinality of the natural exponential function;

[0061] Set the classification threshold to complete obstacle classification.

[0062] In the above implementation, the millimeter-wave radar signal is classified and identified for obstacles through a logistic regression model. First, relevant features k are extracted from the preprocessed signal. These features may include signal amplitude, spectral characteristics, time characteristics, etc. The logistic regression model calculates the linear combination of the input features k through parameters w and b, and maps the results to the probability space p. By setting a classification threshold (for example, 0.5), when p is higher than the threshold, the signal is classified as a specific obstacle type; otherwise it is classified into other categories. This method can perform preliminary classification of radar signal samples through simple and efficient calculations, and is suitable for intelligent traffic scenarios with high real-time requirements. The logistic regression model has a small amount of computation and can quickly complete signal classification. It is suitable for intelligent transportation applications with high real-time requirements. By optimizing the model parameters w and b, the logistic regression model can adapt to a variety of signal characteristics, thereby improving classification accuracy and applicability. The adjustability of the classification threshold enables the model to flexibly adjust the classification criteria according to the needs of different scenarios. For example, in highly sensitive scenarios, the threshold can be lowered to reduce the missed detection rate. The logistic regression model can be used as a preprocessing step for complex models (such as deep learning) to provide input signal screening after preliminary classification, further optimizing system performance.

[0063] S4 includes defining the priority scoring index of obstacle information and calculating the priority score of each obstacle. The specific formula is: S = m1×d+m2×u+m3×v;

[0064] Among them, S represents the comprehensive score of the obstacle, m1 represents the distance weight, m2 represents the size weight, m3 represents the speed weight, d represents the distance between the obstacle and the millimeter-wave radar, u represents the size of the obstacle, and v represents the speed of the obstacle;

[0065] The obstacles are sorted according to the scores and corresponding handling strategies are generated.

[0066] In the above implementation, the obstacle priority score is calculated by a weighted model. Specifically, the comprehensive score S is a weighted linear combination of three key factors: distance, size, and speed. Among them:

[0067] d represents the distance between the obstacle and the millimeter-wave radar, and is usually given a higher weight m1 to give priority to obstacles that are closer;

[0068] u represents the size of the obstacle, and the weight m2 reflects the degree of attention paid to obstacles with larger volumes;

[0069] v represents the speed of the obstacle, and the weight m3 is used to emphasize the importance of fast-moving obstacles.

[0070] By setting reasonable weight parameters m1, m2, and m3, the model can dynamically adjust the scoring rules according to actual needs. For example, in highway scenarios, the speed of obstacles can be given priority; in congested urban roads, more attention can be paid to distance factors. According to the calculated score S, all obstacles are sorted to generate a prioritized list. Subsequently, a corresponding processing strategy is formulated for each obstacle according to the priority, such as issuing a warning signal, adjusting the vehicle speed, or planning an obstacle avoidance route. By adjusting the weight parameters in the scoring model, it can adapt to the needs of different scenarios, such as the differences in environments such as urban traffic and highways. Combining the three aspects of distance, size, and speed, the model can provide a more comprehensive priority evaluation, thereby enhancing the accuracy and rationality of obstacle processing, giving priority to high-scoring obstacles (such as fast-approaching objects), which can effectively reduce potential collision risks and improve the safety performance of intelligent transportation systems. The scoring and sorting process has low computational complexity, and can quickly generate priority lists and processing strategies to meet real-time requirements.

[0071] The obstacles are sorted according to the scores and the corresponding processing strategies are generated, including setting a threshold Q. When S>Q, an alarm is triggered or the vehicle path is adjusted. When S≤Q, only the information is recorded or no processing is done.

[0072] In the above implementation, by defining a threshold Q of the priority score S, obstacles are divided into two categories according to the importance of the score: high-priority obstacles (S>Q) and low-priority obstacles (S≤Q). For high-priority obstacles, the corresponding processing strategy is triggered, such as: issuing an alarm to alert the driver or system to potential dangers; automatically adjusting the vehicle path to avoid obstacles to reduce the risk of collision. For low-priority obstacles, only the information is recorded for subsequent analysis, or no further processing is chosen to reduce the waste of computing resources. This classification and processing method can effectively balance the real-time requirements and resource consumption of the system, while ensuring that timely and effective response measures are taken for high-risk obstacles.

[0073] By setting the threshold Q, high-risk and low-risk obstacles can be quickly distinguished, making the processing strategy more targeted. Immediate action can be taken for high-priority obstacles, while low-priority obstacles are only recorded or ignored, greatly reducing the real-time processing burden of the system. For obstacles with S>Q, alarms are triggered or the path is automatically adjusted, which significantly improves the vehicle's ability to respond to emergencies and enhances driving safety. By screening and processing only high-priority obstacles in real time, it reduces system resource waste and improves computing efficiency and hardware utilization.

[0074] Setting a classification threshold to complete obstacle classification includes: when p>0.5, it is judged as a vehicle; when p≤0.5, it is judged as a pedestrian. In the above implementation, the probability value p output by the logistic regression model is combined with the set classification threshold of 0.5 to classify the target detected by the millimeter wave radar: when p>0.5, the model believes that the characteristics of the target are closer to the characteristics of the vehicle, such as larger size, faster speed, etc., so the target is judged to be a vehicle; when p≤0.5, the characteristics of the target are closer to the characteristics of pedestrians, such as smaller size and slower movement speed, so the target is judged to be a pedestrian. This classification method is based on the linear calculation results of the logistic regression model, maps complex target characteristics into a simple probability value p, and achieves rapid distinction between vehicles and pedestrians by setting a fixed threshold (0.5), providing a basis for subsequent traffic behavior prediction and system response. The probability output and fixed threshold of the logistic regression model are used to achieve fast classification, which simplifies the target recognition process in complex scenarios. The classification process has low computational complexity and is suitable for traffic scenarios with high real-time requirements, such as autonomous driving and intelligent traffic management. By adjusting the weights and bias items of the logistic regression model, the classification performance can be further optimized, and misjudgments and missed judgments can be reduced. The classification model can be expanded according to the needs of different scenarios, such as adding more categories (such as non-motor vehicles) or optimizing the feature extraction process.

[0075] S3 also includes processing the pre-processed signal features through a dimensionality reduction algorithm, and the specific formula is: r′=Y×r; wherein r′ represents the eigenvector after dimensionality reduction, Y represents the dimensionality reduction matrix, and r represents the original eigenvector.

[0076] In the above implementation, the preprocessed signal features are processed by a dimensionality reduction algorithm to reduce redundant information and improve computational efficiency. The specific implementation is as follows: r is a multidimensional original feature vector extracted from the millimeter wave radar signal, which contains various attribute information of the signal, such as spectrum characteristics, amplitude changes, and time series characteristics;

[0077] Y is a dimension reduction matrix generated by a dimension reduction algorithm (such as principal component analysis PCA or linear discriminant analysis LDA). The dimension of this matrix is ​​determined by the dimension of the target feature space, and the components that best represent the original feature information are usually selected by optimization;

[0078] Through matrix multiplication r′=Y×r, we get the reduced eigenvector r′, which has a significantly reduced dimension but retains the most representative information. The reduced eigenvector will be used as the input of subsequent classification algorithms (such as logistic regression or deep learning) to further improve the efficiency and accuracy of target recognition and classification;

[0079] The dimensionality reduction process effectively avoids the interference of noise and redundant information in the original high-dimensional feature vector on classification and recognition performance, while reducing the computational complexity, making it suitable for intelligent transportation scenarios with high real-time requirements.

[0080] By reducing the dimension of the feature space through dimensionality reduction, the computational complexity of subsequent classification and recognition algorithms is reduced, and the real-time performance of the system is improved. The dimensionality reduction process removes redundant information and noise, making the retained features more concentrated and representative, thereby improving the accuracy of classification and recognition. After dimensionality reduction, the amount of feature data is reduced, saving storage and transmission resources. It is suitable for resource-constrained embedded systems. The dimensionality reduction algorithm has a wide range of applicability, and different methods (such as PCA, LDA or nonlinear dimensionality reduction methods) can be selected according to target requirements to meet diverse feature extraction and processing needs.

[0081] The operating frequency range of the millimeter wave radar is 24 GHz to 77 GHz. In the above embodiment, the operating frequency range of the millimeter wave radar is set to 24 GHz to 77 GHz to meet the diverse needs in the intelligent traffic scene: 24 GHz and 77 GHz are the main operating frequency bands of the current millimeter wave radar. Among them, the 24 GHz frequency band has a larger coverage range and is suitable for detecting targets at a longer distance; while the 77 GHz frequency band has higher resolution and accuracy, which is suitable for identifying close or small targets; this frequency range coverage enables the millimeter wave radar to meet the needs of long-distance detection and high-precision identification at the same time. In traffic scenes, low frequency bands (such as 24 GHz) are suitable for scenes such as highways that require long-distance detection, while high frequency bands (such as 77 GHz) are suitable for accurate detection of close-range obstacles in complex urban environments. By reasonably designing the operating frequency range of the radar, spectrum interference with communication equipment in other frequency bands (such as Wi-Fi and mobile communications) can be effectively avoided to ensure signal stability and reliability. The operating frequency range covers 24 GHz to 77 GHz, so that the millimeter wave radar has both long-distance detection capabilities and high-precision close-range identification capabilities to meet the multiple needs of the intelligent transportation system. Different frequencies are adapted to different traffic scenarios, which enhances the applicability of millimeter-wave radar in a variety of environments. For example, low-frequency bands are suitable for large-scale obstacle warnings, and high-frequency bands are suitable for accurate identification of close-range dynamic targets. Reasonable frequency selection can reduce signal interference, improve the robustness of the radar system, and provide more reliable data support for detection in complex traffic environments. In specific scenarios, specific frequency bands can be selectively activated according to needs, thereby optimizing power consumption and improving the energy efficiency of radar equipment.

[0082] S4 also includes generating obstacle information, including the type, position and motion state of the obstacle. The motion state is determined by calculating the acceleration of the obstacle. The specific formula is: a=(l t+1 -l t ) / Δt;

[0083] Where a represents the acceleration of the obstacle, l t+1 represents the speed of the obstacle at time t+1, l t represents the speed of the obstacle at time t, t represents time, and Δt represents the time interval.

[0084] In the above implementation, comprehensive obstacle information including type, position and motion state is generated by real-time tracking of the position and speed of obstacles: the type of obstacle (e.g., vehicle, pedestrian, non-motor vehicle, etc.) is determined based on the output of a classification model (e.g., logistic regression or deep learning algorithm). The three-dimensional coordinates of the obstacle relative to the radar are calculated based on the target reflection position data of the millimeter-wave radar signal to determine its precise position. The speed of the obstacle at different time points is obtained by analyzing the Doppler effect signal of the millimeter-wave radar. t and l t+1 According to the formula a=(l t+1 -l t ) / Δt, and get the acceleration information of the obstacle. Acceleration describes the rate of change of the target speed and is an important parameter of the target's motion state. The speed and acceleration of the obstacle are combined to describe its motion trend (such as acceleration, deceleration or uniform motion). By calculating the acceleration, not only the current speed of the obstacle can be described, but also its future motion trend can be predicted, providing key data support for the decision-making of the intelligent transportation system, integrating the type, location and motion state of the obstacle into one, and providing more comprehensive information for subsequent path planning and obstacle avoidance strategy generation. Based on the high-frequency data acquisition and signal processing of millimeter-wave radar, the motion state can be calculated in real time to meet the high real-time requirements of the intelligent transportation system. Whether it is a fast-moving vehicle or a slow-moving pedestrian, this method can improve the adaptability of the system in complex traffic environments through accurate description of the motion state.

[0085] A millimeter wave radar obstacle detection system for intelligent transportation is also provided, which is used to implement the steps of the millimeter wave radar obstacle detection method for intelligent transportation, and the system includes:

[0086] A radar signal acquisition module is used to obtain the original radar signal collected by the millimeter-wave radar;

[0087] A signal preprocessing module connected to the radar signal acquisition module, used to preprocess the original radar signal based on a multipath reflection model to filter out false targets caused by reflections from high-rise buildings;

[0088] The obstacle recognition and classification module connected to the signal preprocessing module is used to use the deep learning algorithm to perform obstacle recognition and classification on the preprocessed radar signal;

[0089] The information generation module connected to the obstacle recognition and classification module is used to generate obstacle information according to the recognition and classification results.

[0090] The workflow of the entire system is as follows:

[0091] Signal acquisition: The radar signal acquisition module receives the original signal of the millimeter-wave radar and transmits it to the signal preprocessing module in real time.

[0092] Signal preprocessing: The signal preprocessing module uses a multipath reflection model to decompose and filter the original signal, filter out false targets caused by environmental reflections and extract valid signals.

[0093] Identification and classification: The obstacle recognition and classification module uses a deep learning model to classify the preprocessed signals, determines the target type based on the classification probability p, and outputs the classification results.

[0094] Information generation: The information generation module generates information such as the type, location, and motion status of obstacles based on the comprehensive classification results, and generates corresponding processing strategies based on priority scores and threshold judgments.

[0095] By processing radar signals in modules, the system has high flexibility and scalability, and the functions of each module are clear, which is easy to optimize and upgrade. The modules work together to realize real-time processing, classification and information generation of radar signals, meeting the time-sensitive needs of intelligent transportation systems. The generated obstacle information covers multiple dimensions such as type, position, speed, acceleration, etc., providing complete support for the decision-making of intelligent transportation systems. The preprocessing module based on the multipath reflection model effectively filters out false targets caused by environmental reflections, improving the accuracy of obstacle detection. Through parameterized configuration, it can adapt to the detection needs of various traffic scenarios such as highways and urban roads.

[0096] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A millimeter wave radar obstacle detection method for intelligent transportation, characterized in that: The method comprises: S1. Obtaining the original radar signal collected by the millimeter wave radar; S2. Based on the multipath reflection model, the original radar signal is preprocessed to filter out false targets caused by reflections from high-rise buildings, including decomposing the signal into long-term trends, seasonal components, and noise residuals, extracting the trend component as the true target signal, removing the high-frequency seasonal component, and outputting the signal to reduce false targets caused by reflection interference. The specific formula for decomposing the signal is: y[n]=T[n]+S[n]+R[n]; Where y[n] represents the original signal collected by the millimeter-wave radar, T[n] represents the long-term trend component in the signal, S[n] represents the seasonal component in the signal, R[n] represents the noise residual in the signal, and n represents the sampling index of the millimeter-wave radar collected signal; The specific formula of the output signal is: z[n]=y[n]-S[n]; Wherein, z[n] represents the effective signal after preprocessing, y[n] represents the original signal collected by the millimeter-wave radar, S[n] represents the seasonal component in the signal, and n represents the sampling index of the millimeter-wave radar collected signal; S3, using deep learning algorithms to identify and classify obstacles on preprocessed radar signals; S4. Generate obstacle information based on the recognition and classification results.

2. The millimeter wave radar obstacle detection method for intelligent transportation according to claim 1, characterized in that: The S1 includes: Get the original radar signal and design a bandpass filter to extract the target frequency band in the signal. The specific formula is: g(t) = x(t) × h(t); Among them, g(t) represents the target signal obtained after filtering, x(t) represents the original signal collected by the millimeter-wave radar, h(t) represents the characteristic function of the bandpass filter, and t represents time.

3. The millimeter wave radar obstacle detection method for intelligent transportation according to claim 1, characterized in that: The S3 includes: Define the classification model. For each radar signal sample, use the trained logistic regression model to predict the probability of it belonging to a certain category. The specific formula is: Where p represents the probability value of the millimeter-wave radar signal being classified as a certain type of obstacle, w represents the weight parameter vector of the model, k represents the preprocessed feature extracted from the millimeter-wave radar signal, b represents the bias term, and e represents the cardinality of the natural exponential function; Set the classification threshold to complete obstacle classification.

4. The millimeter wave radar obstacle detection method for intelligent transportation according to claim 1, characterized in that: The S4 includes: Define the priority scoring index of obstacle information and calculate the priority score of each obstacle. The specific formula is: S = m1×d+m2×u+m3×v; Among them, S represents the comprehensive score of the obstacle, m1 represents the distance weight, m2 represents the size weight, m3 represents the speed weight, d represents the distance between the obstacle and the millimeter-wave radar, u represents the size of the obstacle, and v represents the speed of the obstacle; The obstacles are sorted according to the scores and corresponding handling strategies are generated.

5. The millimeter wave radar obstacle detection method for intelligent transportation according to claim 4, characterized in that: The obstacles are sorted according to the scores and the corresponding processing strategies are generated, including setting a threshold Q, triggering an alarm or adjusting the vehicle path when S>Q, and only recording information or not processing when S≤Q.

6. The millimeter wave radar obstacle detection method for intelligent transportation according to claim 3, characterized in that: The setting of the classification threshold to complete obstacle classification includes: when p>0.5, it is judged as a vehicle, and when p≤0.5, it is judged as a pedestrian.

7. The millimeter wave radar obstacle detection method for intelligent transportation according to claim 1, characterized in that: The S3 further includes: The preprocessed signal features are processed by dimensionality reduction algorithm, the specific formula is: r′=Y×r; Among them, r′ represents the eigenvector after dimensionality reduction, Y represents the dimensionality reduction matrix, and r represents the original eigenvector.

8. The millimeter wave radar obstacle detection method for intelligent transportation according to claim 1, characterized in that: The operating frequency range of the millimeter wave radar is 24 GHz to 77 GHz.

9. The millimeter wave radar obstacle detection method for intelligent transportation according to claim 1, characterized in that: The S4 further comprises: When generating obstacle information, it includes the type, position and motion state of the obstacle. The motion state is determined by calculating the acceleration of the obstacle. The specific formula is: a = (l t+1 -l t ) / Δt; Where a represents the acceleration of the obstacle, l t+1 represents the speed of the obstacle at time t+1, l t represents the speed of the obstacle at time t, t represents time, and Δt represents the time interval.

10. A millimeter wave radar obstacle detection system for intelligent transportation, used to implement the steps of the millimeter wave radar obstacle detection method for intelligent transportation according to any one of claims 1 to 9, characterized in that: The system comprises: A radar signal acquisition module is used to obtain the original radar signal collected by the millimeter-wave radar; A signal preprocessing module connected to the radar signal acquisition module, used to preprocess the original radar signal based on a multipath reflection model to filter out false targets caused by reflections from high-rise buildings; The obstacle recognition and classification module connected to the signal preprocessing module is used to use the deep learning algorithm to perform obstacle recognition and classification on the preprocessed radar signal; The information generation module connected to the obstacle recognition and classification module is used to generate obstacle information according to the recognition and classification results.

Images