Parameters selection algorithm for LS-SVM based on adaptive genetic algorithm and application on camera calibration

Abstract

The precondition of accurate visual system is accurate camera calibration. Used Least Squares Support Vector Machines (LS-SVM) can achieve the camera calibration. The nuclear function parameter and penalty parameter is a pivotal factor which decides performance of LS-SVM. Usually, most users select parameters for an LS-SVM by rule of thumb, so they frequently fail to generate the optimal approaching effect for the function. In order to get optimal parameters automatically, a new approach based on an adaptive genetic algorithm (AGA) is presented, which automatically adjusts the parameters for LS-SVM, this method selects crossover probability and mutation probability according to the fitness values of the object function, therefore reduces the convergence time and improves the precision of GA, insuring the accuracy of parameter selection. This method was applied to Camera calibration, and simulation results showed the validity to improving the calibration accuracy.