Automatic parking obstacle sensing method based on millimeter wave radar

Abstract

The invention belongs to the technical field of automobile automatic parking, and particularly relates to an automatic parking obstacle sensing method based on millimeter wave radar, a de-noising network and a classification network are finally obtained through data acquisition, network model construction and model training, the de-noising network can output de-noised target data according to an RDMAP input in real time and a corresponding working temperature, and the classification network is used for classifying the target data according to the RDMAP input in real time. The position of the target relative to the radar can be calculated through cfu, speed measurement, angle measurement and distance interpolation, and the category of the target can be output through multi-frame RDMAP accumulation and a classification network; the denoising network eliminates noise energy, and the target distance precision can be improved; besides, the temperature of the radar radio frequency chip is detected in the data set making and real-time processing process and is used as one of network model inputs, the robustness of the denoising network at different temperatures is improved, and noise residues caused by the temperatures are avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of automotive automatic parking, and particularly relates to an automatic parking obstacle perception method based on a millimeter-wave radar. Background Art

[0002] To avoid safety accidents, during the automatic parking process, the intelligent driving system must always pay attention to the surrounding environment of the vehicle body. As one of the "organs" for perception of the intelligent driving system, the millimeter-wave radar must have the ability to accurately perceive the environment. Currently, most applications of millimeter-wave radars are during driving, while ultrasonic radars are mostly used for environment perception during parking. The vehicle-mounted millimeter-wave radar adopts a frequency-modulated continuous-wave system, and the sampling starting distance is zero. In theory, the distance blind zone is extremely small, so it still has a role to play in detecting near-range targets. In fact, due to the influence of DC signal noise, both the near-range target detection performance and the ranging performance are affected. Considering the missed alarm rate and false alarm rate, the near-range target detection ability is greatly reduced.

[0003] The patent "An Automatic Parking Method and System Based on 4D Millimeter-Wave Radar" mentions using a 4D millimeter-wave radar to detect the reference distances of the vehicle tail from various reference objects, without considering the actual near-range target detection performance and ranging accuracy of the millimeter-wave radar; the paper "An Obstacle Detection Method for a Parking System" and the patent "An Automatic Parking Obstacle Distance Detection Method, Its System, and Detection Equipment" both use multiple ultrasonic sensors for obstacle ranging, with relatively high ranging accuracy. However, ultrasonic radars can only output distance information and do not have information such as target azimuth, height, and speed. The patent "A Vehicle-Mounted Millimeter-Wave Radar Target Classification Method, Device, System, and Storage Medium" uses the energy distribution diagram in the Doppler frequency-time dimension as the network input for target classification. However, during the signal preprocessing process, the Doppler dimension stationary clutter is filtered out by vector mean cancellation. When the radar is stationary or moving at a low speed, important targets will be lost. Especially in the parking mode, most targets are stationary or low-speed targets relative to the radar. Moreover, this patent uses a single-frame image for classification, with a relatively low accuracy. The paper "Human Detection Based on Time-Varying Signature on Range-Doppler Diagram Using Deep Neural Networks" uses multiple-frame images as inputs, with an improved classification accuracy, but has a large computational amount and occupies a large amount of memory, and is not suitable for vehicle-mounted devices. Summary of the Invention

[0004] The present invention mainly proposes an automatic parking obstacle perception method based on a millimeter-wave radar for the automatic parking scenario, to meet the requirement of the intelligent driving system for accurately perceiving obstacles around the vehicle body during parking, so as to achieve efficient and safe automatic parking.

[0005] The vehicle-mounted millimeter-wave radar transmits a linear frequency modulated continuous wave signal:

[0006] in is the fast time, T is the duration of a single chirp (which can be understood as a pulse) (i.e., pulse width), is the carrier frequency, is the fast time frequency, and correspond, is the modulation frequency. rect represents the rectangular function, and the expression is , exp is the exponential function, the expression is , j represents the imaginary part.

[0007] The target reflected echo is:

[0008] in is the distance from the target to the radar array, and c is the speed of light. After receiving the echo, the frequency modulation is processed by demultiplying the echo by conjugation with the reference signal to obtain a single-frequency signal, and , that is, the frequency of the single-frequency signal is proportional to the distance of the target. The echo is processed by deinterlacing and Fourier transform to obtain a one-dimensional range image, and each sampling frequency point corresponds to a target at a different distance.

[0009] The vehicle-mounted radar has a short operating range, and the target echo delay at the maximum operating range is also smaller than the pulse width. Therefore, the sampling starting distance is zero, and the corresponding starting frequency is also zero.

[0010] The sampling start frequency is zero, which will inevitably introduce DC and low-frequency noise into the echo. When the noise power is too large, it will drown out the radar's close-range targets and create a detection blind spot. The detection blind spot varies with the radar's distance resolution. The worse the resolution, the larger the detection blind spot. In the automatic parking scenario, it is necessary to accurately perceive the environment around the vehicle body to avoid collisions.

[0011] In order to solve the problem of radar close-range detection blind spots, the present invention proposes an automatic parking obstacle perception method based on millimeter-wave radar, and the specific steps are as follows: Step S1: No target dataset creation, Collect the 2D FFT results of the distances of each receiving channel of the radar in a variety of non-target scenarios (non-target scenarios include various open scenarios where the radar is stationary, various open scenarios where the radar is moving at low speed, and scenarios where the radar is moving at high speed. Among them, the low-speed movement speed of the radar is less than 2 m / s, and the high-speed movement speed of the radar is greater than 10 m / s). Take out the data of multiple range-dimensional sampling units in the near-range area of the radar for each receiving channel (the near-range area of the radar refers to the area within 0 - 5 meters from the radar, and this near-range area of the radar can be adjusted according to actual applications). While collecting the data, record and take out the temperature of the radar RF chip to ensure that the non-target data set covers as many scenarios as possible and includes all working temperatures (upper and lower limits) that the radar RF chip may encounter. Finally, the taken-out data forms a non-target data sample.

[0012] It should be noted that the range dimension refers to the dimension in the direction of radar wave propagation, which is used to measure the straight-line distance between the target and the radar; the unit is: the actual distance represented by each sampling point. The sampling starting distance is the radar array surface, and the space is sampled according to the resolution. It is the working temperature of the radar RF chip, not the ambient temperature. The radar generates heat during operation, causing the internal temperature to rise, which is to some extent affected by the ambient temperature, mainly depending on the radar power and heat dissipation performance. The radar should consider the ambient temperature during design. Therefore, the above-mentioned working temperature of the radar RF chip is all working temperatures considering extreme environmental conditions.

[0013] Step S2: Target data set production Set various targets in the near-range area of the radar, and classify the targets into stationary strong targets, stationary weak targets, and moving targets according to the target characteristics. Among them, the moving targets are set as targets with a relative movement speed exceeding one speed resolution unit of the radar. The corresponding stationary targets are set as targets with a relative movement speed not exceeding one speed resolution unit of the radar. The stationary targets are further divided into strong targets and weak targets, mainly distinguished by the scattering intensity of electromagnetic waves. The stationary strong targets are set as strong-scattering stationary targets, including but not limited to metals and walls; the stationary weak targets are set as weak-scattering stationary targets, including but not limited to cones and green belts.

[0014] Multiple distance angles are set for each target, and the 2D FFT results of the distances of each receiving channel of the radar are collected under the radar motion states of the same target (the radar motion states are divided into a stationary state and a low-speed motion state, and the speed of the radar during low-speed motion is lower than 2 m / s). The data of multiple distance-dimensional sampling units in the near-range area of the radar for each receiving channel are taken out. While collecting the data, the temperature of the radar RF chip is recorded and taken out, so that the data taken out forms a target data sample, and the set target is labeled, and the label corresponds to the target category.

[0015] Step S3: Construct a denoising network. Use the target-free data sample in Step S1 as the training sample, input it into the denoising network for training, and output the residual. Continuously train and iterate until the residual is equal to zero, so as to obtain a denoising network with a residual equal to zero.

[0016] To avoid the influence of DC and low-frequency noise on the detection of near-range targets by the radar, that is, to eliminate the DC and low-frequency noise in the echo. In Step S1, data of multiple target-free scenarios are collected (the target-free scenarios include multiple empty scenarios where the radar is stationary, multiple empty scenarios where the radar is moving at low speed, and the radar high-speed motion scenario). Since there must be no targets in the near-range area of the radar in the empty scenario and the radar high-speed motion scenario (the radar high-speed motion scenario is to collect noise samples under high-speed motion for better denoising processing), the 2D FFT data collected at this time is the DC and low-frequency noise, and this noise will change with temperature. Therefore, the temperature of the radar RF chip is input at the same time. Through training, when the above data is input, the network output residual is close to zero (theoretically, the residual is equal to zero). This is the purpose and expected result of the training, so that the denoising network remembers the characteristics of this noise.

[0017] Step S4: Use the target data sample in Step S2 as the training sample, input it into the denoising network finally obtained in Step S3 for training, and output the residual. Perform target detection on the residual result to determine whether there is a target, and obtain the target detection position and target category. Continuously train and iterate until the target detection position and its target category correspond one by one to the target position and its target category set in Step S2, so as to obtain a target detection network (that is, the trained denoising network) and a classification network.

[0018] The 2D FFT data collected in Step S2 is the DC and low-frequency noise + target data. Through training, the denoising network can remove the noise data in this type of data according to the noise characteristics and retain the target data, so as to obtain a target detection network (that is, the trained denoising network) and a classification network.

[0019] In summary, when the input data is an empty scene or a high-speed moving scene, the residual is close to zero, indicating no target; when the input data is the target echo, the residual is the target energy after denoising. Perform target detection (CFAR) on the residual result, perform range interpolation, and select appropriate antenna array elements for angle measurement to obtain the target position.

[0020] In steps S1 and S2, when retrieving the data of multiple range-dimensional sampling units in the near-range area of the radar for each receiving channel, if the data volume is small, a lightweight network can be selected for training to facilitate on-chip deployment.

[0021] Step S5: Denoise, detect, and identify target real-time data When the radar is working normally, collect the 2DFFT results of the ranges of each receiving channel of the radar under normal real-time operation. At this time, there is data in the near-range area of the radar (the near-range area of the radar refers to the area within 0 - 5 meters from the radar) and data in the far-range area of the radar (the far-range area of the radar refers to the area more than 5 meters from the radar). The data in the far-range area of the radar can be used for the radar function during normal driving of the vehicle; the data in the near-range area of the radar can be used for the radar function during vehicle parking.

[0022] When the radar moving speed is higher than 2 m / s, the near-range target detection module of the radar is not enabled for near-range target detection.

[0023] When the radar moving speed is lower than 2 m / s, the near-range target detection module of the radar is enabled for near-range target detection. Near-range target detection includes: retrieving the data of multiple range-dimensional sampling units in the near-range area of the radar for each receiving channel. At the same time, record and retrieve the real-time temperature of the radar RF chip, so that the retrieved data forms target real-time data. Input the target real-time data into the target detection network finally obtained in step S4. After passing through the denoising network, target data for each channel is obtained. Sum the data for each channel to obtain the range-Doppler map (RDMAP). Perform target detection on the range-Doppler map, and then obtain the target information for the current frame through range interpolation, speed measurement, and angle measurement (that is, perform conventional signal processing on the RDMAP: CFAR (target detection), speed measurement, angle measurement (DOA), range interpolation, to obtain the target position information). Accumulate multiple frames of range-Doppler maps (RDMAP), and retrieve three consecutive frames of range-Doppler maps and input them into the classification network to obtain the target category. At this time, output the target category and the saved latest target information for system decision-making.

[0024] If no target is detected in three consecutive frames of range-Doppler maps within the near-range target detection range of the radar, it means there is no target within the radar observation range during parking, then clear the target parameter information and range-Doppler map for the corresponding frames.

[0025] It should be noted that the range-Doppler map is a two-dimensional spectrogram generated by 2DFFT, which reflects the joint distribution of the range and velocity of the target.

[0026] Compared with the prior art, the present invention has the following technical effects: In view of the problem of the near-range detection blind area of millimeter-wave radar, the present invention proposes an automatic parking obstacle perception method based on millimeter-wave radar. Through data acquisition, network model construction, and model training, a denoising network and a classification network are finally obtained. The denoising network can output the denoised target data according to the real-time input RDMAP and the corresponding operating temperature. After cfar, speed measurement, angle measurement, and distance interpolation, the position of the target relative to the radar can be calculated. Through the accumulation of multiple frames of RDMAP, the classification network can output the target category; the denoising network eliminates the noise energy and can also improve the target distance accuracy; in addition, during the production of the data set and the real-time processing process, the temperature of the radar RF chip is detected and used as one of the inputs of the network model, which improves the robustness of the denoising network at different temperatures and avoids noise residues caused by temperature; when there are obstacles within the radar observation range, the target category, position, and speed information can be output, which is convenient for the system to make decisions; when there are no obstacles, false alarms will not occur, avoiding incorrect decisions by the system.

[0027] The present invention will be further described below in conjunction with the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required to be used in the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0029] Figure 1 is a schematic flow chart of the method of the present invention; Figure 2 is a schematic diagram showing the change of the DC noise energy of the second distance sampling unit with the chip temperature in the embodiment of the present invention; Figure 3 is a schematic diagram showing the change of the target energy with the chip temperature in the embodiment of the present invention; Figure 4 is a schematic diagram of the denoising network in the embodiment of the present invention; Figure 5 is a schematic diagram of the classification network in the embodiment of the present invention; Figure 6 is a schematic diagram of the original RDMAP of the second distance sampling unit without a target in the embodiment of the present invention; Figure 7It is a schematic diagram of the denoised RDMAP when there is no target in the second distance sampling unit in the embodiment of the present invention; Figure 8 It is a schematic diagram of the original RDMAP when there is a target in the second distance sampling unit in the embodiment of the present invention; Figure 9 It is a schematic diagram of the denoised RDMAP when there is a target in the second distance sampling unit in the embodiment of the present invention; Figure 10 It is a schematic diagram of the denoised cfar result when there is a target in the second distance sampling unit in the embodiment of the present invention. Detailed implementation manners

[0030] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will describe the detailed implementation manners of the present invention in conjunction with the accompanying drawings. Many specific details are set forth in the following description to fully understand the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0031] For the automatic parking scenario, this embodiment proposes an automatic parking obstacle perception method based on a millimeter-wave radar, which meets the requirement of the intelligent driving system to accurately perceive obstacles around the vehicle body during parking, thereby realizing efficient and safe automatic parking.

[0032] As Figure 1 shown, the specific steps are as follows: (1) Making a no-target data set, Collect the 2DFFT results of the distances of each receiving channel of the radar in various no-target scenarios (the no-target scenarios include various empty scenarios where the radar is stationary, various empty scenarios where the radar moves at a low speed, and the radar high-speed movement scenario, where the speed of the radar moving at a low speed is less than 2 m / s, and the speed of the radar moving at a high speed is higher than 10 m / s). Take out the data of multiple distance-dimensional sampling units in the near-range area of the radar for each receiving channel (the near-range area of the radar refers to the area within 0-5 meters from the radar, and this near-range area of the radar can be adjusted according to actual applications). While collecting the data, record and take out the temperature of the radar RF chip to ensure that the no-target data set covers as many scenarios as possible and includes all possible operating temperatures (upper and lower limits) of the radar RF chip. Finally, the taken-out data forms a no-target data sample.

[0033] To better illustrate that the energy of the DC and low-frequency noise changes with the temperature of the radar RF chip, this embodiment takes out the data of the second distance sampling unit (the second distance sampling unit is in the distance dimension) in the near-range area of the radar, as Figure 2As shown in the figure, it is a schematic diagram of the DC noise energy of the second distance sampling unit changing with the chip temperature of the radar RF chip. It can be concluded that the DC noise energy of the second distance sampling unit decreases with the increase of the chip temperature of the radar RF chip.

[0034] (2) Target dataset production Set various targets in the near range of the radar, and classify the targets into stationary strong targets, stationary weak targets, and moving targets according to the target characteristics. Multiple distance angles are set for each target, and the 2DFFT results of the distances of each receiving channel of the radar in the radar motion state (the radar motion state is divided into a stationary state and a low-speed motion state, and the speed of the radar during low-speed motion is lower than 2 m / s) of the same target are collected. The data of multiple distance-dimensional sampling units in the near range of the radar for each receiving channel are taken out. While collecting the data, the temperature of the radar RF chip is recorded and taken out, so that the data taken out forms a target data sample, and the set targets are labeled, and the labels correspond to the target categories.

[0035] To better illustrate the change of the target energy with the temperature of the radar RF chip, the target energy at 50 m is taken out in this embodiment. As Figure 3 shown in the figure, it is a schematic diagram of the target energy changing with the chip temperature. It can be concluded that the target energy at 50 m decreases with the increase of the chip temperature of the radar RF chip.

[0036] In summary, both the noise energy and the target energy are related to the operating temperature of the radar RF chip. Therefore, the corresponding operating temperature of the radar RF chip needs to be recorded during data collection, and the operating temperature of the radar RF chip is used as an input at the same time, which can avoid missed detection and false detection caused by temperature changes.

[0037] (3) Network model construction First, a denoising network is constructed. The target-free data samples are used as training samples and input into the denoising network for training, and the residuals are output. The training is continuously iterated until the residuals are equal to zero, so as to obtain a denoising network with residuals equal to zero (i.e., a lightweight denoising network).

[0038] In this embodiment, the 2DFFT results of the second distance sampling units (the second distance sampling units are in the distance dimension) in the near range of the radar for each receiving channel are taken out as inputs, arranged in the Doppler dimension and the channel dimension to obtain a two-dimensional image. The U-net is selected as the denoising network and pruned, and finally a lightweight denoising network is obtained, as Figure 4 shown in the figure.

[0039] Then, a target detection network is constructed: The target data samples are used as training samples and input into the lightweight denoising network for training, and the residuals are output. Target detection is performed on the residual results to determine whether there is a target, and the target detection position and target category are obtained. Continuous training iterations are carried out until the target detection position and its target category correspond one by one to the set target position and its target category, thereby obtaining the target detection network and the classification network.

[0040] After the denoising network is trained, a target detection network is constructed. After passing through the denoising network, the real target echo can be obtained. Multiple frames are accumulated as the input of the target detection network, and the target detection network is trained. The output result is the target category, corresponding to the label obtained by classification according to characteristics. In this embodiment, the range-Doppler map (RDMAP) obtained by accumulating 3 frames is used as the input, a convolutional neural network is selected as the feature extraction network, and a fully connected layer is used as the classifier to obtain a fully connected neural network (i.e., the classification network), as Figure 5 shown.

[0041] (4) Real-time denoising, detection, and recognition of target data When the radar is working normally, the 2DFFT results of the distances of each receiving channel of the radar under normal real-time working conditions are collected. At this time, there are data in the near-range area of the radar (the near-range area of the radar refers to the area within 0 - 5 meters from the radar) and data in the far-range area of the radar (the far-range area of the radar refers to the area more than 5 meters from the radar). The data in the far-range area of the radar can be used for the radar function during normal driving of the vehicle; the data in the near-range area of the radar can be used for the radar function during vehicle parking.

[0042] When the radar moving speed is higher than 2 m / s, the near-range target detection module of the radar is not enabled for near-range target detection.

[0043] When the radar moving speed is lower than 2 m / s, the near-range target detection module of the radar is enabled for near-range target detection. The near-range target detection includes: taking out the data of multiple range-dimensional sampling units in the near-range area of each receiving channel of the radar. At the same time, the real-time temperature of the radar RF chip is recorded and taken out. Thus, the data taken out forms the target real-time data. The target real-time data is input into the target detection network. After passing through the denoising network, the target data of each channel is obtained. The data of each channel is summed to obtain the range-Doppler map (RDMAP). Target detection is performed on the range-Doppler map, and then current frame target information is obtained through range interpolation, speed measurement, and angle measurement (that is, conventional signal processing is performed on the RDMAP: cfar (target detection), speed measurement, angle measurement (doa), range interpolation, to obtain target position information). Multiple frames of range-Doppler maps (RDMAP) are accumulated, and 3 consecutive frames of range-Doppler maps are taken out and input into the classification network to obtain the target category. At this time, the target category and the latest saved target information are output for system decision-making.

[0044] If no target is detected in three consecutive frames of range-Doppler maps within the radar's short-range target detection range, it indicates that there is no target within the radar's observation range during parking. In this case, clear the target parameter information and range-Doppler maps for the corresponding frames.

[0045] In this embodiment, take the 2DFFT results of the second distance sampling units (the second distance sampling units are in the range dimension) in the short-range areas of each receiving channel as inputs. As Figure 6 shown, it is a schematic diagram of the original RDMAP when there is no target in the second distance sampling unit; as Figure 7 shown, it is a schematic diagram of the denoised RDMAP when there is no target in the second distance sampling unit; as Figure 8 shown, it is a schematic diagram of the original RDMAP when there is a target in the second distance sampling unit, where the target is a pedestrian; as Figure 9 shown, it is a schematic diagram of the denoised RDMAP when there is a target in the second distance sampling unit, where the target is a pedestrian; as Figure 10 shown, it is a schematic diagram of the denoised cfar result when there is a target in the second distance sampling unit, where the target is a pedestrian.

[0046] In summary, an automatic parking obstacle perception method based on millimeter-wave radar proposed by the present invention uses a denoising network to eliminate the energy of DC signals. This not only improves the detection performance for weak targets but also improves the ranging accuracy of short-range targets. At the same time, during the production of the dataset and the real-time processing process, the working temperature of the radar RF chip is detected and used as one of the network inputs, improving the robustness of the denoising network at different temperatures and avoiding noise residues caused by temperature. After denoising, target detection, speed measurement, angle measurement, and range interpolation are performed to obtain target parameter information. By accumulating multiple frames of energy distribution maps and completing target classification through a classification network. Different from general radar target classification, the present invention directly uses the energy distribution map in the Doppler frequency-time dimension (i.e., the range-Doppler map) as the network input for target classification. The input contains more target feature information, significantly improving the short-range target detection performance and ranging accuracy of millimeter-wave radar. Compared with ultrasonic radars, it can not only output high-precision ranging results but also output the azimuth, altitude, and speed information of the target. In the present invention, only short-range targets are classified, and only the Doppler results of some distance dimension sampling units are selected as network inputs. Similarly, multi-frame accumulation is used, greatly reducing the computational complexity and memory consumption.

[0047] The above are only the preferred embodiments of the present invention and do not impose any formal limitations on the present invention. Any person skilled in the art can, without departing from the scope of the technical solution of the present invention, make many possible changes and modifications to the technical solution of the present invention by using the methods and technical contents disclosed above, or modify it into equivalent embodiments with equivalent changes. Therefore, all equivalent changes made according to the shape, structure and principle of the present invention without departing from the content of the technical solution of the present invention shall be covered by the protection scope of the present invention.

Claims

1. An automatic parking obstacle perception method based on millimeter-wave radar, characterized in that, It includes the following steps: Step S1: Production of non-target data set Collect the 2D FFT results of the distances of each receiving channel of the radar in various non-target scenarios, extract the data of multiple range-dimensional sampling units in the near-range area of the radar for each receiving channel. At the same time, record and extract the temperature of the radar RF chip, and the extracted data forms a non-target data sample. Step S2: Production of target data set Set various targets in the near-range area of the radar. For each target, set multiple range angles. Collect the 2D FFT results of the distances of each receiving channel of the radar in the moving state of the radar for the same target, extract the data of multiple range-dimensional sampling units in the near-range area of the radar for each receiving channel. At the same time, record and extract the temperature of the radar RF chip, and the extracted data forms a target data sample. Label the set targets, and the label corresponds to the target category. Step S3: Construct a denoising network. Use the non-target data sample in Step S1 as the training sample, and continuously train and iterate to obtain a denoising network with a residual equal to zero. Step S4: Use the target data sample in Step S2 as the training sample and input it into the denoising network finally obtained in Step S3, and continuously train and iterate until the target detection position and its target category correspond one by one to the target position and its target category set in Step S2, thereby obtaining a target detection network and a classification network. Step S5: Denoising, detection, and recognition of target real-time data Collect the 2D FFT results of the distances of each receiving channel of the radar during normal real-time operation of the radar, extract the data of multiple range-dimensional sampling units in the near-range area of the radar for each receiving channel. At the same time, record and extract the real-time temperature of the radar RF chip, and the extracted data forms target real-time data. Input the target real-time data into the target detection network finally obtained in Step S4 to obtain the target information of the current frame, and input it into the classification network to obtain the target category.

2. The automatic parking obstacle perception method based on millimeter-wave radar according to claim 1, characterized in that, In Step S1, the non-target scenarios include various empty scenarios where the radar is stationary, various empty scenarios where the radar moves at a low speed, and the radar high-speed movement scenario.

3. The automatic parking obstacle perception method based on millimeter-wave radar according to claim 1, characterized in that, In Step S2, the motion states of the radar are divided into a stationary state and a low-speed motion state. When the radar moves at a low speed, the speed is lower than 2 m / s.

4. The automatic parking obstacle perception method based on millimeter-wave radar according to claim 1, characterized in that, In Step S2, after setting various targets in the near-range area of the radar, the targets are divided into stationary strong targets, stationary weak targets, and moving targets according to the target characteristics.

5. The automatic parking obstacle perception method based on millimeter-wave radar according to claim 1, characterized in that In Step S3, during the process of obtaining a denoising network with a residual equal to zero, it includes: constructing a denoising network, using the non-target data sample in Step S1 as the training sample, inputting it into the denoising network for training, and outputting the residual. Continuously train and iterate until the residual is equal to zero, thereby obtaining a denoising network with a residual equal to zero.

6. The automatic parking obstacle perception method based on millimeter-wave radar according to claim 1, characterized in that In Step S4, during the process of obtaining the target detection position and its target category, it includes: using the target data sample in Step S2 as the training sample, inputting it into the denoising network finally obtained in Step S3 for training, and outputting the residual. Perform target detection on the residual result to determine whether there is a target, and obtain the target detection position and target category.

7. The automatic parking obstacle perception method based on millimeter-wave radar according to claim 1, characterized in that In step S5, in the process of obtaining the current frame target information and the target category, it includes: inputting the target real-time data into the target detection network finally obtained in step S4 to obtain a range-Doppler map, performing target detection on the range-Doppler map, and then obtaining the current frame target information through range interpolation, speed measurement, and angle measurement. Take out consecutive multiple frames of range-Doppler maps and input them into the classification network to obtain the target category.

8. The automatic parking obstacle perception method based on millimeter-wave radar according to claim 1, characterized in that In step S5, when the radar moving speed is higher than 2 m / s, the radar near-range target detection module is not enabled for near-range target detection; When the radar moving speed is lower than 2 m / s, the radar near-range target detection module is enabled for near-range target detection. The near-range target detection includes: taking out the data of multiple range-dimensional sampling units in the radar near-range area of each receiving channel. At the same time, record and take out the real-time temperature of the radar RF chip, so that the data taken out forms target real-time data. Input the target real-time data into the target detection network finally obtained in step S4 to obtain a range-Doppler map, perform target detection on the range-Doppler map, and then obtain the current frame target information through range interpolation, speed measurement, and angle measurement. Take out consecutive multiple frames of range-Doppler maps and input them into the classification network to obtain the target category.

9. The automatic parking obstacle perception method based on millimeter-wave radar according to claim 8, characterized in that If no target is detected in three consecutive frames of range-Doppler maps within the radar near-range target detection range, clear the target parameter information and range-Doppler maps of the corresponding frames.

10. A method for automatically parking obstacle perception based on a millimeter-wave radar according to claim 4, characterized in that, In step S2, a moving target is set as a target with a relative moving speed exceeding one speed resolution unit of the radar; a stationary target is set as a target with a relative moving speed not exceeding one speed resolution unit of the radar; a stationary strong target is set as a strong scattering stationary target, which includes but is not limited to metals and walls; a stationary weak target is set as a weak scattering stationary target, which includes but is not limited to cone barrels and green belts.

Images