Hand Gesture Recognition Research Based On Neural Networks

As a natural way of communication, hand gesture is gradually a new way of human-computer interaction. Hand gesture includes dynamic gesture and static gesture. The staticgesture plays an important role in understanding the dynamic gesture. The first step of handgesture recognition is to separate the hand from the image, and then extract the feature ofhand, and make the recognition according to the different rules and algorithm. This paperdescribes the key technology of hand gesture recognition, and focus on the way of staticgesture segmentation and gesture recognition.As the first step of gesture recognition, the effect of gesture segmentation influencedirectly the precision of recognition. This paper introduces some way of gesture segmentation,makes a deep research about gesture segmentation based on the skin color and compares theresults of gesture segmentation in the different color space. Thereby, this paper presents away of gesture segmentation combined with adaptive background deduction and severalcolor space. The tests manifest that by this way, the gesture segmentation can be finishedeffectively with a good performance in the complicated backgroundIn the way of gesture recognition, this paper implements the algorithm based on NeuralNetworks by deep research. The algorithm based on BPNN has the following advantages:good self-learning and self-organization, parallel processing structure, high accuracy ofrecognition. But its training network time is long, and it has a large calculating quantity andbad extensibility. Since neural network algorithm has some deficiency, this paper studies anew neural network--ELM, and also presents a gesture recognition algorithm based on theELM. The tests prove that this algorithm is an excellent way of gesture recognition which iseasy to achieve, with a quick test speed and high rate of identification.Finally, the paper implements the hardware platform of gesture recognition based onFPGA and DSP. This hardware system adopts the modular design method, and it has highflexibility and strong commonality, and easy to transplant gesture recognition technology tomobile devices.