Detection of human faces and hands using colour and range data
Matjaž Divjak, Damjan Zazula · 2005
The basis for all high-level human-computer interaction applications is a reliable tracking of human body. In this paper we present a method for non-intrusive, video-based detection and tracking of two of the most important body parts: hands and head. It combines colour and range information acquired by a stereocamera. Skin coloured regions are extracted by segmentation based on colour histogram comparison. Range data is partitioned into several depth intervals by observing its homogeneity. Partial results are fused by assuming that objects are present in both modalities. Additionally, 2D coordinates and motion trajectories of objects are determined. The method is evaluated on real-world video sequences and detailed performance estimation is carried out. With sensitivity at 80 %, specificity at 98 % and recognition rate at 95 %, the method shows great potential for home video-based tracking solutions.