Abstract
This paper describes a system for counting persons in the bicycle path in sequences of RGB images acquired by a stationary camera. Skin color is used as one important feature for detecting passing persons from the colored images. One part of the image, not the whole one, is taken as the detection area to decrease the computation complexity and satisfy the need of real-time processing. In order to get the final number of persons, three levels of the output signal from the raw data are obtained: single pulse, one pulse group and two adjacent pulse groups. Lastly, Experimental results based on outdoor scenes have shown an acceptable efficiency.