Moving object detection from moving camera sequences

Abstract

Moving object detection from moving camera sequences is an important subject in field of computer vision. This paper presents a new approach for moving object detection in sequences taken from moving camera, key idea of which is to compensate the camera motion by estimate affine transformation parameters of the background using a combined method of Scale Invariant Feature Transform (SIFT) algorithm and Random Sample Consensus (RANSAC) algorithm. Feature points are detected in consecutive frames by SIFT detector and matched according to Euclidean measure, which is the initial matching step. In order to eliminate incorrect feature correspondences and the correctly matched features in the image region of moving object, RANSAC algorithm is applied to rectify the initial matching results and the affine transformation parameters are estimated accurately. Followed by inter-frame difference and morphology operations, moving object is detected successfully. Tracking of features is robust by using SIFT and the computational complexity is significantly reduced by performing the RANSAC estimation algorithm. The effectiveness of the proposed method is demonstrated using real video sequences from moving cameras.