Human body part detection in RGB-D image with pattern of depth difference and spatial depth difference features

Human body part detection has been an important research topic in the last decade. It is widely applicable in areas such as human activity recognition, pose detection and other applications related to human movements. The objective of a human body part detection system is to associate a body part to each human pixel. Recent studies show that applying depth maps significantly improves the results of body part detection. In this study, two new features based on pixel depth difference is proposed. First feature is based on pixel-wise depth difference between the input pixel and neighbor pixels selected using a weighted circular distribution. The second feature is the difference between coefficients of polynomials fitted to neighbors of the input pixel at difference scales, making the feature invariant scaling. Random decision forest was used for pixel classification. Comparison of results with the state of the art methods reveal that the proposed method is able to distinguish and differentiate the various components of the body more accurately.