LAUSR

The classification of moving vehicles on roadways is an important application of intelligent transportation systems. For successful vehicle classification using radar, suitable feature extraction with high accuracy from the returned echoes of moving vehicles is necessary. In this paper, a novel framework for predicting the length of each moving vehicle is presented based on two different scenarios of the frequency-modulated continuous-waveform (FMCW) radar, i.e., the radar is stationary or moving. Different clutter reduction techniques were considered depending on the motion of the FMCW radar. Using clutter-free signals, moving vehicles are separated on a range-angle map, rather than on a range-Doppler (R-D) map, enabling the successful separation whether the vehicles are overlapping or not on the R-D map. Finally, the scattering center information across multiple frames, rather than a single frame, is fused to significantly increase the accuracy of the predicted length. The proposed framework was demonstrated through experiments in an actual road environment using a commercial FMCW radar. The results indicated a good match with the true lengths of the vehicles.