Robust algorithm for motion analysis based on Active Tubes

Abstract

Determining motion of objects is a very important and difficult problem. Many researches have been studied in this field. We previously presented a new model, named Active Tubes, to find motion of non-rigid objects from an image sequence. Active Tubes analyzes the temporal context in a spatio-temporal solid using an energy minimizing model like Snakes. You can consider it as a kind of accumulation of Snakes along the time axis. In the former paper, we used greedy algorithm to converge Active Tubes. In, this paper, we point out greedy algorithm's weakness to noise, and propose an extended algorithm to deform Active Tubes. The new algorithm is as fast as greedy algorithm and more robust from noise.<>