A target recognition system and method

Abstract

The application discloses a target recognition system and a recognition method. The system comprises: a millimeter wave radar arranged at a traffic intersection and configured to detect intersection vehicle information; a data processing module connected to the millimeter wave radar and configured to process radar echo signals; a signal mixing module configured to mix transmission signals and echo signals of the millimeter wave radar to obtain original signals processed by the radar; a data transformation module configured to perform two Fourier transformations on intermediate frequency signals in a distance dimension and a speed dimension to obtain distance and speed information of a target, output a distance-Doppler image; a data monitoring module configured to continue to detect the intermediate frequency signals and perform a Fourier transformation in an angle dimension to obtain angle information of the target, combine the distance information to obtain a 2D millimeter wave radar point cloud image of the target, and an identification module configured to complete classification and identification of a vehicle model based on a preset classification model. The method realizes accurate classification of vehicles.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of radar digital signal processing, in particular, to a target recognition system and a recognition method. BACKGROUND

[0002] In recent years, with the increase of private car ownership, the traffic environment of road intersections is gradually complex, which promotes the continuous development of the concept of intelligent traffic management in China. The scheduling order and signal cycle of the existing traffic intersection light management system are fixed and cannot change, which cannot cope with the increasingly complex traffic congestion situation. By real-time and accurate monitoring of the intersection, timely adjustment of the signal light passing time according to the on-site situation can greatly reduce the road pressure and improve the passing efficiency. The millimeter wave radar is not affected by weather and light and can truly achieve continuous detection all day long. In addition, the millimeter wave radar has the advantages of long detection distance, high precision, good stability and easy installation.

[0003] At present, millimeter wave radar is widely used in advanced auxiliary driving system for sensing the environment of the traffic vehicle. The road side millimeter wave radar is installed at the traffic intersection, and the targets collected by the millimeter wave radar information acquisition are mainly the distance, speed, angle, trajectory, traffic flow and queue length of the vehicle, and the classification recognition of the traffic vehicle is less concerned, especially the vehicle recognition research based on 77GHz millimeter wave radar. In the literature [Fang Feifei, Yu Wen. Doppler radar vehicle recognition algorithm based on PCA-LDA-SVM [J]. Data Acquisition and Processing, 2012, 27(01): 111-116.], the authors use 24GHz Doppler radar for classification and recognition, and extract the vehicle length, maximum echo energy and other training classification network through machine learning. In the literature [Y. Jin, L. Dong, H. Lu, X. Pan and W. Chang, "Research on vehicle recognition technology based on multi-feature-SVM method," IET International Radar Conference (IET IRC 2020), 2020, pp. 597-603, doi: 10.1049 / icp.2021.0764.], the authors start to use 77GHz millimeter wave radar for experiment, but only divide the vehicle into large and small vehicles, which is not detailed enough. SUMMARY

[0004] In view of the above-mentioned defects, the target identification system based on the millimeter wave radar (77GHz millimeter wave radar) is proposed to simulate the roadside scene to classify the vehicles running on the road surface, and the classification is more detailed.

[0005] To achieve the above-mentioned purpose, the application adopts the following scheme:

[0006] A target identification system has:

[0007] The millimeter wave radar is arranged at the traffic intersection and is used to detect the vehicle information of the lane corresponding to the intersection;

[0008] The data processing module is connected to the millimeter wave radar, and the radar echo signal is obtained by processing through the data processing module;

[0009] The signal mixing module is used to mix the transmission signal and the echo signal of the millimeter wave radar to obtain the original signal processed by the radar;

[0010] The data transformation module is used to perform two Fourier transforms on the intermediate frequency signal in the distance dimension and the speed dimension to obtain the distance and speed information of the target, and output the range-doppler image (RDM);

[0011] The data monitoring module is used to continue to detect the intermediate frequency signal and perform Fourier transform in the angle dimension to obtain the angle information of the target, and combine the distance information to obtain the 2D millimeter wave radar point cloud image of the target;

[0012] The identification module is based on a preset classification model to complete the classification and identification of the vehicle model. The target identification system can reliably and safely simulate the roadside millimeter wave radar scene, accurately simulate the mathematical model when the vehicle is detected by the road test millimeter wave radar, and realize the classification and identification of the vehicle.

[0013] Preferably, the target identification system further comprises a low-pass filter device, the input end of which is connected to the output end of the signal mixing module, and the output end of which is connected to the input end of the data transformation module, which is used to filter the signal after mixing the transmission signal and the echo signal of the millimeter wave radar.

[0014] Preferably, the data processing module is based on modeling of the millimeter wave radar signal and combines the automobile radar scattering cross section to calculate the radar echo signal.

[0015] The application also provides a target identification method based on the above-mentioned target identification system,

[0016] Comprising the following steps:

[0017] S1, obtaining the radar original signal and transmitting it to the data transformation module;

[0018] S2, the data transformation module pre-processes the received original signal to obtain RDM, and detects the intermediate frequency signal based on the data monitoring module, and performs Fourier transform on the angle dimension to obtain the angle information of the target, and combines the distance information to obtain the 2D millimeter wave radar point cloud of the target;

[0019] S3, according to the recognition module based on the received RDM and 2D millimeter wave radar point cloud, the classification and recognition of different vehicle targets are carried out.

[0020] Preferably, the step S3 comprises: after data set pre-processing, inputting a convolutional neural network for classification network training to complete the classification and recognition of different vehicle targets.

[0021] Preferably, the step S1 comprises:

[0022] The millimeter wave radar transmitting signal is mathematically modeled, the automobile radar scattering cross section calculated by the simulation tool is combined, the radar echo signal is calculated, the intermediate frequency signal, i.e. the original signal of the radar processing, is obtained by mixing the transmitting signal and the echo signal.

[0023] Preferably, the step S2 comprises:

[0024] The intermediate frequency signal is subjected to two times of Fourier transform on the distance dimension and the speed dimension to obtain the distance and speed information of the target, and output RDM; and the intermediate frequency signal is subjected to constant false alarm rate detection, and then subjected to Fourier transform on the angle dimension to obtain the angle information of the target, and combine the distance information to obtain the 2D millimeter wave radar point cloud of the target.

[0025] Preferably, the traffic intersection scene, the radar transmitting signal and the target radar scattering cross section are simulated or simulated.

[0026] Preferably, the mathematical expressions of the millimeter wave radar transmitting signal, the echo signal and the intermediate frequency signal in the above embodiment are as follows

[0027]

[0028]

[0029]

[0030] wherein, x T (t) is the transmitting signal, x R (t) is the echo signal, x IF (t) is the intermediate frequency signal, f0 is the starting frequency of the frequency-modulated continuous wave, μ is the slope of the frequency-modulated continuous wave, is the initial phase, P is the energy at the receiving antenna derived from the radar equation, RCS is the radar cross section area of the vehicle, τ(t) is the time difference of the radar from transmission to reception, j represents the imaginary unit, and exp represents the exponential function.

[0031] Advantages

[0032] Compared with the existing vehicle identification and classification scheme based on roadside millimeter wave radar, the embodiment of the present application can reduce the time cost and labor cost, realize the performance evaluation of the roadside millimeter wave radar deployment setting, and can distinguish at least four types of vehicles. Experimental results show that the target recognition system proposed in the present application can complete the functions of simulation and collection and classification of vehicle information at the traffic intersection. BRIEF DESCRIPTION OF DRAWINGS

[0033] The present application will be further described below in conjunction with the drawings and embodiments:

[0034] Figure 1 is a schematic block diagram of the target recognition system of the embodiment of the present application;

[0035] Figure 1a is a schematic diagram of the target recognition system of the embodiment of the present application;

[0036] Figure 2 is the RDM of the millimeter wave radar signal processing output of the system of the embodiment of the present application.

[0037] Figure 3 is the point cloud diagram of the millimeter wave radar signal processing output of the embodiment of the present application.

[0038] Figure 4 is the training curve of the deep learning of the automobile classification and identification of the embodiment of the present application. DETAILED DESCRIPTION

[0039] EMBODIMENT

[0040] As Figure 1 shown is a block diagram of the target recognition system proposed in the present application, which comprises:

[0041] a millimeter wave radar configured at a traffic intersection for detecting vehicle information of a lane corresponding to the intersection;

[0042] a data processing module connected to the millimeter wave radar, which obtains radar echo signals through processing by the data processing module, a signal mixing module for mixing the transmission signals and echo signals of the millimeter wave radar to obtain original signals processed by the radar;

[0043] a data transformation module for performing two Fourier transforms of the intermediate frequency signals in the distance dimension and the speed dimension to obtain distance and speed information of the target and output RDM;

[0044] a data monitoring module configured to continue detecting the intermediate frequency signal, and perform a Fourier transform in the angle dimension to obtain angle information of the target, and combine the distance information to obtain a 2D millimeter wave radar point cloud map of the target;

[0045] The recognition module is configured to complete the classification and recognition of the vehicle model based on a preset classification model. The target recognition system is used in a simulation scene of a traffic intersection. In implementation, the millimeter wave radar is arranged at the traffic intersection to detect vehicle information of a lane corresponding to the intersection, and is transmitted back to the data processing module.

[0046] The millimeter wave radar described above has a signal generator connected to a transmitting antenna TX and a mixer, and a receiving antenna RX receives a signal and outputs the signal through the mixer (see Figure 1a ). The signal is transmitted to a data transform module through a low-pass filter connected to the mixer for millimeter wave radar (referred to as radar) signal processing. Then, the classification and recognition of the vehicle are performed by the recognition module. Preferably, the recognition module is connected to a convolutional neural network module, the convolutional neural network module receives the output of the RDMap and outputs to the recognition module after training through the convolutional neural network module, so as to realize the classification and recognition of the vehicle target.

[0047] In an embodiment, the radar in the simulation scene is located on a traffic light pole, about 6 meters from the ground, and detects three lanes, each lane being 3.75 meters wide, with a maximum detection distance of 61.5 meters and a maximum detection speed of 16 meters / second. The target vehicle randomly appears on the three lanes and approaches the traffic intersection at a random speed under the premise of being consistent with reality.

[0048] In classification, a mathematical model of the radar signal is calculated according to the simulation radar signal and the scattering cross-section area of the car radar, and then the target information such as distance, speed and angle is obtained by three times of fast Fourier transform in the distance dimension, the Doppler dimension and the angle dimension, and the RDM (Range-Doppler Map) and the 2D millimeter wave radar point cloud map are output.

[0049] The RDM is preprocessed, a data set is made, a classification network is trained, and the classification and recognition of the vehicle target are realized.

[0050] Next, experiments are performed to verify the effectiveness of the target recognition method of the present application. In the experiments, the above system is used to classify and recognize the vehicle.

[0051] The target recognition method comprises the following steps:

[0052] 1) modeling the millimeter wave radar signal, using the BART simulation tool, inputting the digital model of the car, the azimuth angle, the pitch angle and the sweep frequency range, outputting the frequency response of the scattering cross-section area of the car radar, and after inverse Fourier transform, combining the time domain transmitting signal multiplication, the radar echo signal is obtained mixing to obtain an intermediate frequency signal complete the signal simulation of radar detection process,

[0053] wherein: f0 is the starting frequency of the frequency-modulated continuous wave, μ is the slope of the frequency-modulated continuous wave, is the initial phase, P is the energy at the receiving antenna obtained by the radar equation, RCS is the radar cross section area of the vehicle, τ(t) is the time difference from the radar transmission to the reception.

[0054] 2) The intermediate frequency signal is subjected to two Fourier transforms in the distance dimension and the speed dimension, the distance and speed information of the target is obtained, and the RDM is output.

[0055] 3) Continue to perform constant false alarm rate detection on the intermediate frequency signal, and then perform Fourier transform in the angle dimension to obtain the angle information of the target, and combine the distance information to draw a 2D millimeter wave radar point cloud map of the target.

[0056] 4) Select different types of vehicle models to repeat steps 1) and 2) to obtain a large number of Range-Doppler Maps, and perform data set preprocessing operations such as translation, rotation, and noise addition.

[0057] 5) The data set is subjected to deep learning of the convolutional neural network, the classification network is trained, and the function of classifying and identifying the vehicle model is completed.

[0058] Referring to Figure 2 and Figure 3 , the preset condition is that a small car is in the first lane, i.e. the lateral distance is 2 meters, and is 40 meters away from the radar, and moves towards the intersection at a speed of 10 meters per second.

[0059] Figure 2 is the range-velocity map of radar signal processing, and the highlighted position is the detected target, which contains the information that the target is 40 meters away from the radar sensor and the vehicle speed is 10 meters per second.

[0060] Figure 3 is the point cloud map of the target, which shows the information that the target vehicle is 2 meters away from the radar sensor in the lateral direction and 40 meters away in the longitudinal direction.

[0061] Referring to Figure 4 , four types of vehicles are selected in an embodiment of the application, which are:

[0062] Bus, trolleybus, car, motorcycle. The state of the vehicle is constantly adjusted, i.e. the distance of the vehicle to the radar sensor (25m~40m), the speed (0.5m / s~10m / s) and the lane (1, 2, 3) in which the vehicle is located, to simulate the corresponding distance-velocity diagram output. In this embodiment, the four types of vehicles are adjusted for 20 times respectively, and finally 80 distance-velocity diagrams are obtained. The data set is expanded to 1120 images, the size is adjusted to the input size of the convolutional network 451x451x1, and after seven convolutional layers and maximum pooling downsampling layers, the output size is 7x7x512. After the full connection layer, the whole steps of the convolutional neural network are completed. Through iterative training of the classification network, vehicle classification and recognition are realized. The results are shown in Figure 4 Figure 4 The training curve when training the network), and the verification success rate of vehicle recognition is 92.81%.

[0063] From the experimental results, it can be seen that the target recognition system and method proposed in the present application can complete radar signal simulation and vehicle classification and recognition in a simulated scene.

[0064] The above embodiments are only for illustrating the technical concept and characteristics of the present application, and the purpose is to enable those skilled in the art to understand the content of the present application and implement it, and cannot limit the protection scope of the present application. Any equivalent transformation or modification made according to the spirit and essence of the present application should be covered within the protection scope of the present application.​

Claims

1. A target identification method for a target identification system, the target identification system comprising: a 77GHz millimeter wave radar configured at a traffic intersection for detecting vehicle information of a lane corresponding to the traffic intersection. A data processing module is connected to the millimeter wave radar, and radar echo signals are obtained through the data processing module; a signal mixing module is used to mix the transmission signals and echo signals of the millimeter wave radar; and a low-pass filter device is used to filter the signals after the transmission signals and echo signals of the millimeter wave radar are mixed; A data transformation module is used to perform two Fourier transformations in the distance dimension and the speed dimension on the intermediate frequency signals to obtain distance and speed information of the target and output a range-Doppler map (RDM); a data monitoring module is used to continue to detect the intermediate frequency signals and perform a Fourier transformation in the angle dimension to obtain angle information of the target, and combine the distance information to obtain a 2D millimeter wave radar point cloud map of the target; and an identification module comprises the following steps: S1, obtaining radar original signals and transmitting the radar original signals to the data transformation module; S2, the data transformation module performs a preprocessing stage on the received original signals to obtain an RDM, and based on the data monitoring module, performs a constant false alarm rate detection on the intermediate frequency signals and then performs a Fourier transformation in the angle dimension to obtain angle information of the target, and combines the distance information to obtain a 2D millimeter wave radar point cloud map of the target; S3, according to the identification module, performing classification network training based on the received RDM and 2D millimeter wave radar point cloud map to perform classification and identification of different vehicle targets.

2. The target identification method of claim 1, wherein: In step S3, after data set preprocessing, a convolutional neural network is input to perform classification network training to complete classification and identification of different vehicle targets.

3. The target identification method of claim 1, wherein: Step S1 comprises: The transmission signals of the millimeter wave radar are mathematically modeled, the calculated radar scattering cross section of the automobile is combined with a simulation tool, the radar echo signals are calculated, the transmission signals and echo signals are mixed to obtain intermediate frequency signals, and the intermediate frequency signals are the original signals processed by the radar.

4. The target identification method of claim 1, wherein: The constructed traffic intersection scene, radar transmission signals, and target radar scattering cross section are all simulated or realized by simulation.

5. The target identification method of claim 1, wherein: The mathematical expressions of the millimeter wave radar transmission signals, echo signals, and intermediate frequency signals are as follows where x T (t) is the transmitted signal, x R (t) is the echo signal, x IF (t) is the intermediate frequency signal, f0is the start frequency of the frequency-modulated continuous wave, μ is the slope of the frequency-modulated continuous wave, is the initial phase, P is the energy at the receiving antenna derived from the radar equation, RCS is the radar cross section area of the vehicle, τ(t) is the time difference from transmission to reception of the radar, j denotes the imaginary unit, and exp denotes the exponential function.

Images