Research on Calibration Technology of Monocular Camera Based on Improved BAS Algorithm

Abstract

Aiming at the problem of low calibration accuracy and too slow speed of traditional cameras, a monocular camera calibration method based on improved BAS (Beetle Antennae Search) algorithm is proposed. In order to prevent falling into the local optimum, the basic BAS algorithm is generally set to a relatively large initial step length, and the variable step length coefficient of each iteration is a fixed value and close to 1, which causes the local minimum to be continuously jumped out in the early stage. The search efficiency is low and the convergence is slow. Inspired by the fact that SGD (stochastic gradient descent) can be used to iteratively solve the minimum value of the objective function in machine learning, the basic BAS algorithm is improved, and the variable step coefficient δ in each iteration adopts standard exponential decay. The experimental results show that the improved BAS algorithm converges significantly faster than the basic BAS algorithm, and the search results are better, more stable and reliable.