Fast probabilisitic estimation of egomotion from image intensities

This paper proposes a real-time probabilistic solution to the problem of camera motion estimation in a video sequence. Instead of using explicit tracking of features, it only uses instantaneous image intensity variations without prior estimation of optical flow. We represent the camera motion as a probability density which is constructed from the individual motion densities, estimated from spatio-temporal derivatives, of each pixel of the image. The density is formed by accumulating the contribution of each pixel, making it very robust to local perturbations in the image. A fast algorithm is proposed and experimental results show how real-time motion estimation is possible directly from the image stream with good precision.