Citation

Yasutomi S, Mori H, Kotani S. Rob. Auton. Syst. 1996; 17(1-2): 25-34.

Copyright

(Copyright © 1996, Elsevier Publishing)

DOI

10.1016/0921-8890(95)00059-3

PMID

unavailable

Abstract

A new method of finding pedestrians in a time sequence of visual images has been proposed, and its great practical applicability has been demonstrated. Walkers exhibit periodic motion in a quite different way from other moving objects. This periodicity, or in other words rhyth of walking, is the result of the walking action and can be observed in two kinds of periodic intensity change of the pedestrian's image. So, pedestrians can be found if we look for these periodicities which are quite unique to them. To realize this idea, we estimate the temporal-frequency and the spatial-period of the moving objects. The temporal-frequency is the oscillation of intensity against time. The spatial-period represents the fluctuation of intensity along the trajectory of the object, and is referred to as the stride of walking. A pedestrian detection method using these two oscillations has the following advantages. (1) This method is independent of the absolute intensity or contrast of the objects. (2) Clothes, hairstyle, shape of body, illumination and the distance from the camera do not affect the reliability of detection. Finally, the algorithm is simple enough for real time operation. The detailed algorithm is presented here and its validity and practical applicability are examined in an experiment on asphalt paved road.