Bidirectional Multilane Vehicle Counting Approach Based on Multiscale of Multitraffic States Using MIMO Radar

Due to the excellent performance under any light and weather conditions, radar has become one of the most important sensors for traffic surveillance in intelligent transportation system (ITS). This article proposes a bidirectional multilane vehicle counting approach using frequency-modulated continuous wave (FMCW) multiple-input multiple-output (MIMO) radar. The proposed approach can count vehicles based on multiscale of multitraffic states in a variety of traffic behaviors such as fast passing, overtaking, lane changing, congestion, and even forced stops occurring simultaneously on this road. To adapt the complex traffic situations, we classify the instantaneous traffic state by exploiting traffic features in the current frame. Then, we apply target detection, Doppler estimation, and point cloud clustering. After this, we classify a period of traffic state by several frames. The trajectory tracking and vehicle counting are conducted for the resulting period of traffic states. Specifically, we design a trajectory fusion strategy to reduce the impact of trajectory discontinuity and to ensure that the vehicle trajectory only counted once for the congestion situation. Furthermore, we propose a selection strategy to overcome the multitrajectory segments caused by the fast situation. Experimental results show the improved performance of the proposed approach.