Efficient online spatio-temporal filtering for video event detection

Abstract

We propose a novel spatio-temporal filtering technique to improve the per-pixel prediction map, by leveraging the spatio-temporal smoothness of the video signal. Different from previous techniques that perform spatio-temporal filtering in an offline/batch mode, e.g., through graphical model, our filtering can be implemented online and in realtime, with provable lowest computational complexity. Moreover, it is compatible to any image analysis module that can produce per-pixel map of detection scores or multi-class prediction distributions. For each pixel, our filtering finds the optimal spatio-temporal trajectory in the past frames that has the maximum accumulated detection score. Pixels with small accumulated detection score will be treated as false alarm thus suppressed. To demonstrate the effectiveness of our online spatiotemporal filtering, we perform three video event tasks: salient action discovery, walking pedestrian detection, and sports event detection, all in an online/causal way. The experimental results on the three datasets demonstrate the excellent performances of our filtering scheme when compared with the state-of-the-art methods.