Fundamental Matrix Based Moving Object Detection Using Monocular Camera

Abstract

A typical road environment includes a variety of moving/stopping objects. It is important to detect moving objects for safety driving. However, moving object detection is difficult, as background is also moving when ego-vehicle is driving. In this paper, we introduce the method that detects moving objects using fundamental matrix. Fundamental matrix is calculated from previous and current frames. Feature points are obtained using Harris corner detection in previous frame and correspondence points in current frame are estimated by Lucas-Kanade method. The feature points are evenly selected by considering x-coordinate, y-coordinate and feature magnitude of flow for calculating fundamental matrix using RANSAC. Next, whether each point is inlier or outlier is determined and the points with negative direction is detected. In the simulations, we compare the performances according to the selecting method.