Video based adult and child classification by using body proportion

Abstract

Pedestrian detection in uncontrolled environments is a challenging task. There are various researches for video based pedestrian detection. Moreover, to recognize children and adults in digital platforms is helpful for future applications. For instance, if a Closed Circuit Television(CCTV) camera located on a traffic light detects child who is walking through pedestrian way, system could make some service adjustments. Aim of this article is to detect children and adults separately. We used Haar cascade classifiers for the implementation. We detected head and body of pedestrians. Then we used relative measurements because we cannot get exact height of people from pixels so we applied another method that is proportioning head size to body size of pedestrians. By this technique, we could discriminate children and adults. The results are promising and shows sufficient accuracy.