Segmentation using spatial-feature clustering from image sequences

A method for segmentation of moving objects in outdoor-unconstrained scenes is presented. This work is motivated by a need to monitor the natural environment by using a remote machine vision system to record animal populations. A d´Alembertian of a spatial-temporal Gaussian filter is convolved with a sequence of image frames in order to obtain an image of temporal zero-crossings. Subsequently, the magnitude of the normal visual motion is estimated at each pixel of the spatial-temporal zero-crossings and a spatial motion-based graph-theoretical (GT) clustering is applied. The clustered pixels are backprojected in the original colour image so that we can obtain a colour history image, which is updated for each frame in a sequence. Finally, the output colour history image is processed by a colour-based version of the GT clustering, such that, the segmented object is represented by unimodal colour clusters. This algorithm is very efficient in removing cluttered complex backgrounds and extracting colour features which are subsequently used for recognition