Surveillance video summarisation by jointly applying moving object detection and tracking

摘要

With the growth of massive storage of surveillance video data, it has become imperative to design efficient tools for video content browsing and management. This paper describes an integrative approach for surveillance video summarisation that jointly apply moving object detection and tracking. In the proposed scheme, moving objects are first detected and tracked. The static summarisation is generated to contain some key frames which provide details of the moving objects. The main advantages of our approach include the preservation of important information and economic computational cost. The high performance background modelling with Gaussian mixture model, together with the multi-scale morphological processing, brings together a highly accurate moving object detection tool. The proposed matching criterions for Kalman filtering enhances the tracking accuracy. We experimented with highway surveillance videos and outdoor surveillance videos, demonstrating satisfactory performances.