A fast algorithm for tracking moving objects based on spatio-temporal video segmentation and cluster ensembles

This paper presents a fast algorithm to segment moving objects in video sequences, as the first step of a fast object tracking system. It is based on the detection of level lines to detect closed objects contours in a scene. The detected objects are clustered using a combination of mean shift and ensemble clustering. The proposed method produces a temporal video segmentation in a fraction of the processing time required by comparable state-of-the-art particle-based methods.