Comparision of Computer Vision and Photogrammetric Approaches for Motion Estimation of Object in an Image Sequence

Abstract

3D tracking plays a vital role in 3D applications by enhancing interaction between real and virtual world. We present various real-time 3D motion estimation approaches developed in photogrammetry and computer vision fields and compare their performance. The methods developed in both fields estimates 3D motion of a moving object relative to a camera or equivalently moving camera relative to the object in an image sequence when its corresponding features are known at different times. We reviewed 3D motion models formulated by different methods related to their geometric properties. We implemented four different methods and analyzed their performance results. Comparison from test datasets from image sequences demonstrated that homography based approaches in both fields were more accurate than relative orientation or essential matrix based approaches under noisy situations.