Estimation of rapid-motion for mobile devices using temporal coherence

As the mobile devices become popular, novel interaction methods are required for expanding the application areas. The motion estimation and its applications to user interaction are gaining the attention of computer vision researchers. Recently Hannuksela et al proposed an effective motion estimation algorithm that operates in realtime mode on the mobile platform. The algorithm restricted the search area of the motion of feature block in order to keep the real-time operation, and as a result it could not support the applications that need the rapid motion. The aim of this paper is to capture a rapid motion without extra computation. We propose a method that exploits the previous displacement vector to locate adoptively the search area. Two schemes, Adaptive Searching and Adaptive Confined Searching are described. The experiments compared the Fixed Searching scheme of the original algorithm with two new schemes using synthetic and real images. The results showed that the new schemes could capture the rapid motion while the conventional one could not.