An Improved Target Classification Method for 4-D Radar Sensor Using RAM

The 4-D radar sensor is an important sensor in automotive driving applications, which generate massive sensing data and is easily interfered by the complex traffic environment with noise signals. An improved target classification method for 4-D radar sensor using range-azimuth heatmap (RAM) is proposed in this article to reduce the computation burden and achieve better accuracy in traffic target classification. The first step of this method is to detect the situational curb targets using the reflection characteristics difference in range scope. Then, the curb targets are recognized and eliminated in the RAM using deep learning (DL) method, generating a reduced point cloud set. Consequently, the dynamic clustering is applied to divide the point cloud set into clusters using learning strategy. Finally, a light-weighted network is elaborately designed to extract the locations and categories of the traffic targets on the reduced data. Experiments are carried out to evaluate the proposed method, whose results have proven that this method achieves better performances than other typical methods on localization, classification, and efficiency.