Motion Vector Estimation of Textureless Objects Exploiting Reaction-Diffusion Cellular Automata
Miho Ushida, Alexandre Schmid, Tetsuya Asai, Kazuyoshi Ishimura and Masato Motomura
Conventional motion estimation algorithms extract motion vectors from image sequences based on the image’s local-brightness differences in consecutive images. Therefore, motion vectors are extracted along the moving edges formed by moving objects over their background. However, in the case of “textureless” moving objects, motion vectors inside the objects cannot be detected because no brightness (texture) differences exist inside the object. Severe issues may incur in motion-related imaging applications because motion-vectors of vast (inner) regions of textureless objects can not be detected, although the inner part is moving with the object’s edges. To solve this problem, we propose an unconventional image-processing algorithm that generates spatial textures based on object’s edge information, allowing the detection of the textures motion. The model is represented by a 2-D crossbar array of a 1-D reaction-diffusion (RD) model where 1-D spatial patterns are created inside objects and aggregated to form textures. Computer simulations confirm the approach, showing the formation of textures over approaching objects, which may open applications in machine vision and automated decision systems.
Keywords: Texture generation, cellular automata, reaction-diffusion systems, unconventional image processing