Relative Motion Estimation Based on Sensor Eigenfusion Using a Stereoscopic Vision System and Adaptive Statistical Filtering

This paper presents a method to estimate the relative motion between two vehicles with high accuracy. The estimated quantities are intended to be used as a reference system for automotive sensing techniques and online embedded motion-estimation algorithms. We propose the sensor eigenfusion which makes use of a stereoscopic vision system mounted on-board of a host vehicle. Highly reliable markers, i. e., QR-codes, mounted on a remote vehicle are used for robust features detection and tracking. In the case of the mentioned camera system, the proposed method uses the 3D reconstruction capabilities of stereoscopic vision and optical flow techniques usually used in monocular vision systems. The measurements are then shaped, smoothed and fused using a Kalman filter. To achieve the required high accuracy the characteristic statistical parameters of the filter are adapted online according to confidence measures which depend both on the 3D reconstruction and on the optical flow analysis.