Research And Implementation Of Dynamic Gesture Recognition System Based On

AI as the current hot research in the current society and future for a long period of time will be used in a lot of applications, which you can judge on the popularity of variety of smart furniture and smart single product. And gesture recognition is among the branch of artificial intelligence. Currently, the identification process of dynamic gesture recognition include dynamic gesture segmentation, feature extraction and gesture pattern recognition. The main problem we faced is the recognition speed and accuracy. In order to deal gesture with very fast hand gesture segmentation and feature extraction, the paper by means of high-speed and efficient FPGA platform features to expedite the processing of our dynamic gesture recognition front-end, due to FPGA is not good at high complexity pattern recognition algorithm, so back-end uses the PC host computer to deal with recognition. Value feature transfer between FPGA and PC using serial transmission scheme to achieve.Based on the above background and problems we have to face, this paper presents a dynamic gesture recognition system based on FPGA, and build a prototype for the system. Dynamic gesture segmentation and feature extraction gesture was performed by FPGA, gesture pattern recognition was performed by PC.After dynamic gesture segmentation and feature extraction of FPGA platform, which mainly through skin detection and edge detection to extract the fingertip position. In the skin color detection areas, non-uniform illumination have a large effect on the outcome. This paper presents pixel-level adaptive threshold skin color model to get a robust detection on the basis of the current fixed threshold detection algorithm. In order to improve processing speed and throughput of the system, we use a dual parallel pipelined video stream processing architecture to realize our algorithm.After the FPGA platform to extract the position eigenvalues, characteristic pattern will be sent through the serial port to the PC side of the host computer for pattern recognition. The PC end take advantage of GRT(Gesture Recognition Toolkit) to achieve, which is an open source machine learning library developed by MIT. The reason for using this open source library is that the GRT has advantage like scalability, ease to use, etc. The most important thing is that the GRT has a complete library of machine learning algorithms. This is very important for rapid developing prototype.The number of gesture to be recognized was set to ten, currently. The evaluation of system uses the recall rate, accuracy, precision and confusion matrix to evaluate the reliability of the identification system. After test of the final sample library identification and we found recognition accuracy and real-time systems are just like we expected.