Parametric analysis of KLT algorithm in autonomous driving

Abstract

The Kanade-Lucas-Tomasi(KLT) tracking algorithm is a widely used feature tracking algorithm in the field of computer vision(CV). The selection of proper warping parameters for the estimation of optical flow between adjacent image frames is crucial to obtain accurate and robust tracking results. We compare the various warping parameter settings in an autonomous driving environment based on the modified KLT algorithm with some well-known techniques. The skew and rotation parameters did not show better performance, but rather made convergence more difficult. The scale-parameter-added model has the best performance among the sets of warping parameters.