Video analysis of traffic accidents based on projection extreme learning machine
Xinman Zhang, Tingting He, Longbin Lu, Shuangling Yue, Dongxu Cheng, Xuebin Xu · 2017
Recently, increasing attentions have been concentrated upon the judgement of video-based traffic accidents, which serves as a crucial technology of intelligent transportation system (ITS). This paper introduces a novel vehicle tracking method based on projection extreme learning machine (PELM) by taking consideration of poor performance in the dealing with high-dimensional and low-samples. Firstly, samples near the target in the first frame of a video is employed as training samples to the PELM classifier. The algorithm uses projection vectors instead of random assignment as the weights of the input layer. Then the PELM classifier is trained with the samples near the target in the following frames to update parameters and obtain the target location with the maximum response. Finally, the fitting error and maximum-range trajectory are adopted to judge whether the vehicle is collided or not. Simulation experiments have shown the proposed method can achieve quite satisfying results.