FPGA based real time hand detection by means of colour segmentation

Abstract

Colour is an important feature for object classification in digital video images, where shape-based techniques are not practicable. This can be the case, if real-time processing has to be applied to situations, where many candidate shapes can occur. However, colour can be represented in differ- ent colour spaces. Which of them is appropriate cannot be fixed in general for all purposes. In this paper, the influence of colour space is illustrated by an FPGA based real time hand detection sys- tem. The latter presents a challenging problem, where shape-based approaches are unpromising. This detection system consists of an artificial neural network, which classifies particular single pixels whether they belong to the hand or to the background. Successive use of binary morpho- logical operations improves results.