Miho Ushida: “A self-organizing model of spatial patterns and its evaluation towards motion detection of textureless objects”
CiNet 1F Conference Room
Host: Ferdinand Peper (PI)
If we calculate the motion-vector or optical flow of textureless objects without change, they can be detected by only their outline. This is an important problem for anomaly detection and for the design of interfaces for video games.
To solve this problem, we propose a nonlinear image processing model that generates patterns with a certain spatial frequency from the outlines of textureless objects. This model is arranged in a three-dimensional configuration of one-dimensional networks based on a large number of nonlinear functional units in a crossbar state, and each network performs self-organization of spatial patterns based on a reaction-diffusion model. We show by numerical simulation that the same pattern is generated inside the objects unless they transform during motion.