A Method Based on Dense Trajectory for Violent Video Classification
Nan Wang, Wei Guo Song, Jianjun Hou, Jing Yu · Advances in intelligent systems research/Advances in Intelligent Systems Research · 2016
At present, the internet technology develops so rapidly and the video becomes the major component of the internet traffic.The content security of massive public videos is an important factor to the social stability.Among them, violent video is an important class of unsafe videos.We proposed a novel method based on dense trajectory and extreme learning machine to recognize them.The spatial-temporal characteristics were well expressed by the use of optical flow and gradient.The experiment on the benchmark dataset named Movies indicated our proposed method had a better accuracy than the state-of-the-art methods.Our proposed method is an efficient method for violent video classification.Relative Works DT.Dense Trajectory Feature reflects the continuous changing trend of the motion field in the video.It predicts the change of the position of the sampled pixels by calculating the optical flow field