基于改进差分进化算法的相机标定研究

目前相机标定多采用Tsai和Weng的基于非线性成像模型的分步标定法。在研究标定算法的过程中发现,待优化的目标函数具有多个局部极值点,采用传统的局部优化算法会导致迅速收敛到局部极值点,而非全局最优解,影响了标定精度。差分进化是一种收敛速度很快的进化算法,但当初始种群分布不够理想时,它可能快速收敛到局部极值点。将其改进后用于相机标定,提出了分步法与改进的差分进化算法相结合的相机标定方法。数据实验表明:标定精度比直接采用分步法有所提高。

Camera calibration based on improved differential evolution algorithm

Tsai's and Weng's two-stage methods with distortion model are often used in camera calibration. It is found that objective function has several local extremums, calibration parameters optimized with traditional optimization methods converge to a local extremum. Differential evolution is a new algorithm to get the global optima, its rate of convergence is faster than general evolution algorithm, but when the initial population is not good enough it might converge to local extremum quickly. An improved differential evolution algorithm is proposed, and together with two-stage calibration method, a new camera calibration method is established. Testing results using real data show that it improves calibration accuracy.