Nature inspired optimization techniques for Camera calibration

Nature inspired optimization algorithm are techniques which imitate some natural phenomenon to find an optimum solution of a problem. We have explored the potential of three nature inspired continuous optimization techniques Simulated Annealing, Genetic Algorithms and Particle Swarm Optimization in solving the Camera calibration problem in computer vision. Our experiment setup has demonstrated the calibration accuracy achieved with these algorithms with respect to noise levels, variable bounds and number of control points. Results are compared with Tsai , Zhang and Heikkil calibration accuracy. Stochastic nature of these algorithms yields different results for same experiment settings. In order to gain reliable results we have quantitatively determined the probability of achieving accurate results by applying statistical tolerance interval on fitness values computed by each algorithm for various trials. The statistical evaluation shows that in terms of solution accuracy the PSO surpasses SA and GA and produces better results than their non natural competitors Tsai , Zhang and Heikkil , whereas the probability of achieving a correct solution with a desired fitness value range is high with GA. Our work can serve as a guideline for comparative evaluation of natural optimization algorithms with respect to solution quality, reliability and efficiency.