EFFICIENT CLASSIFICATION OF LOW-RESOLUTION RANGE PROFILES OF AUTOMOBILES USING A COMBINATION OF USEFUL FEATURES

The range proflle (RP) of an automobile is derived by compressing the wideband radar signal, and it can be utilized for the classiflcation and thus contribute to lane change and collision avoidance. However, the limited radar bandwidth due to the cost and the system complexity impedes the successful classiflcation. This paper proposes an e-cient method to construct an e-cient feature vector of the automobile RP through combined use of the central moment, the information on the maximum-minimum and the peak information. Simulation results using the flve automobile models composed of point scatterers and a simple nearest neighbor classifler prove that the proposed method improves the classiflcation result, especially in the multi-aspect classiflcation.