A New Marching Pixels Algorithm for Application-Specific Vision Chips for Fast Detection of Objects' Centroids

D. Fey, C. Gaede, A. Loos, and M. Komann (Germany)


Parallel and Distributed Algorithms, Parallel Computer Ar chitectures, Parallel VLSI Design, Embedded Algorithms


The paper presents a novel distributed and parallel algo rithm for the fast detection of objects’ centroids which is well-suited for an implementation in a vision chip. The algorithm is based on the Organic Computing concept of Marching Pixels (MPs). MPs present a kind of virtual hard ware agents which are crawling in a pixel field to fulfill specific tasks. The MP algorithm is based on the iterative processing of moments that can be mapped very well to the distributed and parallel approach of MPs. Compared to pre vious MP algorithms we developed before, the presented algorithm works in a stronger emergent way leading to a higher processing performance. In the paper we present in detail the mathematical background of the distributed algo rithm. The algorithm was mapped on a fine-grain parallel architecture which was synthesized for an FPGA and an ASIC. The ASIC synthesis proofs the potential of the ap proach for future smart mega pixel CMOS camera chips allowing processing rates up to one thousand frames per second.

Important Links:

Go Back