Evolutionary agents for epipolar geometry estimation

This paper presents an evolutionary agent-based approach to epipolar geometry estimation. Each agent stands for a minimum subset for computing fundamental matrix, and evolves autonomously in the vast solution space to get the optimal result. In so doing, the agents rely on some reactive behaviors such as reproduction and diffusion, and collaborate with others with a subset template. Experimental results show that our approach performs better than other typical methods in terms of accuracy and computational efficiency, and is robust to noise and outliers.