Pedestrian Detection using Fuzzy Clustering and Histogram of Oriented Gradients

Pedestrian detection with complex scenes from a video is an challenging task due to illumination variations and dynamic background changes. In this context, this paper proposes a efficient approach of using fuzzy clustering and Histogram of Gradients for detecting humans. For this task, the paper first initializes a background frame where only static objects are present using fuzzy c-means clustering. Then a background subtraction and foreground detection algorithm (binary threshold) is proposed for detecting the moving objects. A contour is fetched for these foreground objects and passed through HOG classifier for pedestrian detection. The proposed method has been tested on various complex scenes from different data-sets.And it has presented better results over literature methods in terms of classification accuracy of 92%.