Vehicle motion segmentation using rigid motion constraints in traffic video

Abstract

Vehicle motion segmentation is an important task in Intelligent Transportation Systems (ITS) for the sustainable development of smart city. In this paper, a novel trajectory clustering method is proposed using the rigid motion constraints (RMC) for the vehicle motion segmentation. Firstly, camera calibration is done for converting the 2D images to a copy one in 3D world. Then, the inverse projection image (IPI) can be obtained with a known 3D height information and the 2D image. Secondly, we present and verify the RMC of the feature point trajectories in the same rigid object and different rigid objects. By analyzing simulation results, we use the RMC as a measure to define the similarity function. Finally, spectral clustering is adopted to achieve the trajectory clustering, which contains two stages: initial clustering and inter-class merging. By applying this clustering method in actual traffic scene, much more stable clustering results can be obtained. Experimental results on traffic video demonstrate that the proposed method has a good performance in different traffic scenes. Moreover, its accuracy can reach 96% compared with other methods.