Research On Target Azimuth Estimation And Recognition Technology Of Automotive Millimeter Wave Radar

Autonomous driving is one of the development trends of the automobile industry.In recent years,there have been continuous breakthroughs in this field,and even some manufacturers have already produced finished products.Millimeter-wave radars are mainly used in cars for advanced driving assistance systems(ADAS,Advanced Driving Assistance Syatem),and together with other sensors such as lidars and cameras to provide cars with information about the surrounding environment.Compared with other sensors,millimeter wave radar has the advantages of all-day,all-weather and low cost.Because the accuracy of traditional radar is lower than other sensors,the main function of vehicle-mounted radar is still to achieve target detection and follow-up tracking of the target.However,with the development of semiconductors and other related technologies,vehicle-mounted millimeter-wave radars will have higher resolution and will obtain more information about targets.Therefore,this paper mainly studies the related technologies of vehicle-mounted millimeter-wave radar target recognition to make up for the disadvantages of other sensors’ poor target recognition performance in severe weather and climate.Based on the point cloud formed by the vehicle-mounted millimeter-wave radar,this paper realizes the target recognition function of the millimeter-wave radar.It mainly conducts key technologies such as target orientation estimation,radar point cloud clustering and classifier design in the signal processing process of the vehicle-mounted millimeter-wave radar.Research,the main research content is introduced below:(1)Research the related algorithms of millimeter wave radar data processing,and conduct detailed research on the azimuth estimation algorithm among them.First,the principle of the position estimation algorithm is briefly explained,then the performance of several algorithms under a single snapshot condition is simulated and compared,and finally the actual data collected by the millimeter wave radar is used for verification.In addition,the related algorithms for estimating the number of signal sources under this condition are studied,which can realize the low position estimation under the multi-target situation,which lays the foundation for obtaining the millimeter wave radar point cloud.(2)The characteristics of millimeter wave radar point cloud are studied,and a suitable point cloud clustering algorithm is selected based on the characteristics of millimeter wave radar point cloud.First,analyze the characteristics of common target point clouds of common millimeter wave radars,and select appropriate clustering algorithms according to the characteristics of point clouds,then improve the DBSCAN clustering algorithm,and propose a clustering algorithm suitable for radar point clouds.The OPTICS algorithm is studied,and the problem of parameter selection based on density clustering is explained from a more macro perspective.(3)Research on the recognition related algorithms of the target point cloud formed by the millimeter wave radar.First,analyze the millimeter wave radar point cloud and extract the features that reflect the target size and speed characteristics.Then use the neural network and support vector machine method to design the classifier,realize the classification of different target point clouds according to the extracted features,and finally use the convolutional neural network that can automatically extract features directly classifies the obtained point clouds and compares the performance of different algorithms.