Automatic Human Segmentation in Video Using Convex Active Contours

This paper presents a new algorithm for automatic detection and segmentation of humans in video. Our algorithm exploits the robustness and the accuracy of the interactive image segmentation using Convex Active Contours, to segment moving persons, but with an unsupervised manner. Based on a collaborative strategy to cluster a set of extracted Selective Space Time Interest Points, the resulting separated moving clusters are used to initialize automatically the seeds for the segmentation. Experiments show a good performance of our algorithm for human detection and segmentation in video without a user interaction.