Irregular Action Recognition in Court with 3D Residual Network
Song Liu, Qiaolin He, Zhu Wang, Yi‐Fei Pu, Yi Zhang · 2020
Several irregular behaviors, such as smoking, absent and dozing, sometimes occur in the court. To ensure the participants in court in order without irregular behaviors is of importance for a regular trial. However, manually handling such numerous surveillance videos to discriminate and prevent irregular actions is a mission impossible. Recently many effective action recognition algorithms had been proposed, which brings opportunity to overcome the difficulty. We take into account the particularities of irregular actions in court, i.e. stationary background and small range of motion, and thus propose our irregular action recognition method based on 3D residual network that focuses on modeling spatiotemporal information of motion in videos. Experiments show our method can achieve a recognition rate of 82.3% on 13 classes of irregular actions in court.