Determination of Movement Vector by Optical Flow

High Speed Detection of Optical Flow Using Hough Transform

Abstract

The objective of this study is to measure the movement vector of vehicles from images of the ground surface, which are obtained with an on-vehicle camera. The movement vector is calculated from optical flow which is a two-dimensional motion field on the image plane. We consider the optical flow will be useful also for detecting vehicle attitude in the near future. Generally, the optical flow is determined by the spatial gradient method or block-matching method, but these methods take a lot of computation time. In this study, two programs were developed using the Hough transform to reduce the computation time. The accuracy of determining the movement vector and the computation time of the two programs were tested and compared to those of the conventional block-matching method. A two-dimensional actuator was used instead of a vehicle to conduct more accurate and reproducible experiments. One of the developed programs was considered to be better than the other two tested programs. The computation time was about 0.3s and the average ratio of the positional error to the movement distance was about 5.5%.