Motion Based Segmentation with an omnidirectional camera
The system is divided in two parts : an omnidirectional camera to detect independant motion in the scene, and a perspective camera to take pictures at higher resolution in the direction of movement. The two cameras are mounted on a PAN-TILT motor. The picture shows the combinaison of the two images.The detection uses the projective geometry of the omnidirectional camera, and a Bayesian Decision helps grouping the pixels to create zones of independant motion.
The motion based segmentation uses the optical flow as major component. The project contains an adaptation of the Differential method for the optical flow derivation of Lukas and Kanade.