Extraction of Moving Areas in Random-Dot Animation with Parallel Computation

Abstract

The observation of random-dot animations requires the motion detection techniques of moving objects from the background, allowing the boundary of the moving areas to be perceived despite that each frame consists of random-dot patterns with no boundaries. The present study aims to extract the moving parts as well as its moving direction on the random-dot animations based on the local parallel computation of the pixel level. Assuming motion is captured in high frame rate, the variable range of an optical flow is restricted within a finite discretized area of pixels. Then, the system is constructed to select the suitable optical flow among the finite possibilities according to the reaction-diffusion equation dynamics, computed in a distributed manner. This process is one of the regularization methods and is described as the minimization process of the potential functional. Using the computation result of the previous frames, we attempted to reduce the computational iterations and to detect the plural objects moving at different speeds. These effects are demonstrated by computer simulations using actual random-dot animations.