综合表观和运动模式的视频轮椅检测方法

摘要

This paper proposes a novel method to detect wheelchairs in video using the combination of appearance and motion pattern. An appearance based detector is used to detect objects frame by frame, where a cascaded detector is designed using the Haar-like feature and the HOG (histograms of oriented gradients) feature. A motion based classifier is proposed further to distinguish appearance-similar wheelchairs and pedestrians, which usually impedes performance of appearance based detectors. The accuracy and the efficiency of the proposed method were validated by extensive experiments.%提出了一种通过综合利用表观和运动模式从视频中检测轮椅(包括坐轮椅的人)的方法.该方法首先使用两种特征描述轮椅的表观特性,并结合利用基于统计学习得到的分类器快速地排除非目标区域.针对由于直立行人和坐在轮椅人之间的表观相似性,轮椅检测易受直立行人干扰的情况,进一步使用运动信息来区分轮椅和正常行人,降低了虚警.在收集的数据上测试了所提方法的性能,实验结果表明提出的方法能够以接近实时的速度进行检测,并且能够有效地降低正常行人的干扰.