An adapted Lucas-Kanade’s method for optical flow estimation in catadioptric images

The optical flow estimation is one of important problem in computer vision. Differential techniques were used successfully to compute the optical flow in perspective images. Lucas-Kanade is one of the most popular differential method that solve the problem of optical flow by given constrain that motion is locally constant. Even if this method works well for the perspective images, this supposition is less appropriate in the omnidirectional images due to its distortion. In this paper, we propose to use new constraint based on motion model defined for paracatadioptric images. This new constraint will be combined with an adapted neighborhood windows witch are adequate to catadioptric images. We will show in this work that these two hypothesis allows to compute efficiently optical flow from omnidirectional image sequences.