Result images from

B. Heisele, C. Wöhler.
Motion-Based Recognition of Pedestrians.
Int. Conf. on Pattern Recognition, pp. 1325-1330, Brisbane, 1998.
 

Object segmentation is performed by the Colour Cluster Flow (CCF) algorithm using a monocular colour camera. Sequences of regions of interest (ROIs) extracted by the CCF technique are classified by the ATDNN. In the images below, ROIs classified by the ATDNN as pedestrians' legs are marked by black rectangles.