Research On Gesture Recognition Based On BP Neural Network

With the rapid development of Human-Computer Interaction technology, gestures ispaid more and more attention by researchers as an efficient and natural way to interact. Agesture recognition technology based on the method of vision or image processing hassome good advantages of real-time effects and a high recognition rate, it has been widelyused in the intelligent system and other areas, but the surrounding environment light andbackground image obtaining process and even shooting angle dependence restrict thefurther development in the future. With the rapid development of sensor technology, thegesture recognition technology based on the acceleration sensor is becoming a newresearch hotspot. However, the gesture recognition technology based on accelerationsensor is still in its infancy, and it should be further improved in the real-time performanceand recognition rate.On the basis of the research status at home and abroad, a gesture recognition methodbased on BP neural network has been designed in this paper. First of all, gesture datasignals are collected by acceleration sensor (AMI602), and then the gesture data signal issent to a txt file saved to the computer through the serial port.Secondly, against the gesture recognition method designed in this paper and theacceleration signal characteristics, the acceleration data of the gesture is processing, itmainly includes extracting the acceleration signal, converting the hexadecimal data signalinto the decimal acceleration value, removing jump points, the linear smoothing filtering,and detecting the effective gesture data. It can provide the smooth and stable accelerationsignal for the gesture recognition algorithm.Finally, this design mainly researches the algorithm of gesture recognition. It includesthe training of BP neural network and the gesture recognition based on BP neural network.Collect a large number of sample data to train the designed BP neural network, and thenuse the trained BP neural network to recognize the predefined gestures.The system applys the MATLAB environment for software platform, the feasibilityof the algorithm is verified by simulation. In experiments, the average recognition rate of the Arab digital is85.8%, and the rate of five custom single-stroke gestures is87.2%.