Vision-based real-time detection of objects with non-homogeneous color distributions using a fuzzy classifier

This paper proposes a real-time object detection method based on a fast feature extraction method and a fuzzy classifier. In particular, this paper considers the challenging task of detecting an object whose appearance is with multiple and non-uniform color distribution. The proposed detection method is implemented in a real-time object detection system using a pan-tilt-zoom camera. Color histogram obtained from distributions of the appearance of an object on a non-uniformly partitioned hue and saturation (HS) color space is used as a feature vector. An efficient method for histogram extraction during the image scanning process is proposed for real-time implementation. For each search window, only histograms of the non-overlapping parts between two successive windows are computed, which reduces detection time. The classifier used is a fuzzy classifier with support vector learning. Experimental results using a pan-tilt-zoom camera verify efficiency of the feature extraction method.