Estimation of depth and 3D motion parameters of moving objects with multiple stereo images by using Kalman filter

In this paper, use of stereo motion sequences is considered to estimate both three-dimensional (3D) motion and depth of 3D moving points. It is assumed that the objects have small size enough to represent them as single features such as points. To reduce the effect of image measurement noise on the estimation of 3D motion and depth, a time based Kalman filter is applied to filter out some of the noise when a sequence of stereo images are used. An approximated dynamic model is considered based on the assumption that the relative motion in 3D space is slow and smooth enough so that the first derivatives of the 3D position can be considered constant over a sampling period. The Kalman filter predicts the structure at the next time step which can be used to predict the location of features in both image planes and 3D space along with their uncertainties. This allows the tracking system to process only small window regions whose size is determined from its uncertainty. The result of estimation can be used in real tasks, such as obstacle avoidance, by predicting the structure at the next times. The uncertainty of the first order random walk model is estimated by the maximum likelihood technique. Simulation results are presented to show the effectiveness of this method