Hand gesture recognition using Optimized Neural Network Shape Fitting on ARM11

Abstract

Various methods of hand gesture recognition have been proposed in the literature, with high recognition rate. But implementing these methods in embedded system is still challenging since image processing applications needs a high-performance processor. In this paper, a hand gesture recognition system is implemented on a system with an OK6410B board. This board has a processor that runs at 532 MHz, which is relatively high for a small processor. The hand gesture recognition method proposed in this paper is based on the Neural Network Shape Fitting. In this paper we propose some modifications to this method. The modifications were pixel randomizing during the initialization step, addition of several neurons in the iterations, using lookup table for distance measurement and simplification of the finger detection. These modifications yielded a faster processing time (0.95s on the OK6410B) and a higher recognition rate (94.44% using still images as input and 84.53% using live input from a webcam).