A real-time DSP-based hand gesture recognition system

In this paper, a real-time hand gesture recognition system implemented on a TMS320DM642 Digital Signal Processor (DSP) of Texas Instruments (TI) is proposed. The highly effective system achieved by combining specialist algorithms and advanced DSP optimization techniques. The proposed algorithms are composed of face detection and hand gesture recognition together with skin segmentation and hand tracking procedures. Cascade classifiers using AdaBoost algorithm are applied in face detection and hand recognition to accelerate the entire system. In addition, many DSP optimization levels such as floating-point numbers to fixed-point numbers conversion, software pipelining etc. are also exploited. The results show that the proposed design achieves up to 50 640×480 16-bit YCbCr frames per second (fps), which is capable of many real-time applications.