A low power colour-based skin detectors for smart environments

We describe an embedded optical system detecting human skin under a wide range of illuminant conditions. Our attention to such a system is justified by the many applications for which skin detection is needed, e.g. automatic people monitoring and tracking for security reasons or hand gesture recognition for fast and natural human-machine interaction. The presented system consists of a low power RGB sensor connected to an energy efficient microcontroller. The RGB sensor acquires the RGB signal from a region in front of it over a wide dynamic range, converts it in the rg chromaticity space directly on chip and delivers the processed data to the microcontroller. This latter classifies the input signal as skin or non-skin, testing its membership to a skin locus, i.e. to a compact set representing the chromaticities of the human skin tones acquired under several illuminant conditions. The system architecture distributes the computational load of skin detection on both hardware and software, providing a reliable skin detection with a limited energy consumption. This makes the system suitable to be used in smart environments, where energy efficiency is highly desired in order to keep the sensors always ready to receive, process and transmit data without affecting the performance.