Classification of Moving Patterns in Crowds

초록

This paper reports a classification analysis of moving patterns in crowd videos by using a 3 dimensional convolution network for feature extractions with several classification models. For analysis experiments, crowd videos are collected and selected, as those are very similar to the crowd situations interested in Korea, from the previous other studies and some extra videos collected by ourselves from the Internet. According to the experiment results, the deep learning architecture can distinguish each interesting moving pattern from other different patterns up to 73% of the classification accuracy.