Real-time Implementation of Vision-based Unmarked Static Hand Gesture Recognition with Neural Networks based on FPGAs

This paper presents a real-time vision-based unmarked hand sign recognition system with neural network (Self-Organizing Maps) based on FPGA. The system consists of image acquisition module, image preprocessing module, feature extraction module, classification module. It is implemented on a single field-programmable gate arrays (Kintex-7) chip which has a broad developing prospect for its powerful compute ability, compact design, low-power consumption, small physical size, etc. Owning to the hardware embedded implementation, the whole system is very small in physical size and consume low power, which can be applied to many integrated embedded systems. The system can recognition 24 static American Sign Language (ASL), and the performance was verificated via simulation and experiment. The result indicated that the system can accomplish the whole hand recognition task in real time (60 frames per second, which is the utmost speed of the adopted camera) with a high accuracy (93.3%).