Tracking moving objects with co-evolutionary snakes

Abstract

A new symbiotic genetic algorithm (SGA) based active contour model (snake) is proposed to track the B-spline contour of obstacles. It exploits the local control properties of the B-spline to decompose the contour into subcontours and optimizes each subcontour in separate genetic algorithms (GA). Unlike the GA-based snake, an SGA snake can track the obstacle's outline more robustly. Application-specific inter-population genetic operators are introduced to reinforce the symbiotic relationship via migration of genetic material. The use of symbiosis dramatically reduces the combinatorics of the search space, when compared to GAs. Results of tracking objects in real road scenarios demonstrate its robustness to noise and stability of convergence when compared to its GA counterpart.