Motion-based recognition of pedestrians

In this paper we present an algorithm for recogniz ing walking pedestrians in sequences of color images taken from a moving camera The recognition is based on the characteristic motion of the legs of a pedestrian walking parallel to the image plane Each image is seg mented into region like image parts by clustering pixels in a combined color position feature space The pro posed clustering technique implies matching of corre sponding clusters in consecutive frames and therefore allows clusters to be tracked over a sequence of images Based on the observation of clusters over time a two stage classi er extracts those clusters which most likely represent the legs of pedestrians A fast polynomial classi er performs a rough preselection of clusters by evaluating temporal changes of a shape dependent clus ter feature The nal classi cation is done by a time delay neural network TDNN with spatio temporal re ceptive elds