Research On The Recognition Method Of Real-time Gesture Instructions

Since the rapid development of science and technology,various electronic products have widely existed in people's daily life,so interaction through electronic products is more in line with the general trend of technological development.At present,various sensor types of sensing devices are developing rapidly,so various methods of using sensors to identify are gradually entering the lives of the people.However,from the current research results,the gesture command recognition method using sensors still needs to improve the real-time performance and recognition rate and other problems.Aiming at the advantages and disadvantages of sensors,this paper adopts a BP neural network recognition method,and improves this method.First,the data information is collected through the ADXL345 sensor,and then stored in the computer through the serial port.Secondly,in view of the characteristics of using the acceleration sensor,in order to be able to more accurately identify the corresponding gesture command signal,therefore,effective acceleration data information is extracted.The forward and backward incremental comparison method is used to remove the jump points;the experiment is compared with different methods,and a three-point linear smoothing filtering method is proposed to filter the collected data;the stationary state intercept method is used to locate the initial position of the filtered data And terminate the operation of position positioning,so that effective acceleration data can be intercepted.Third,a method for extracting the key points of the dispersion extreme value is proposed,and then normalized to complete the extraction of valid data information and efficient compression of data information.Traditional gesture instruction recognition technology often requires too much calculation,and this The operation greatly reduces the adverse effects caused by this problem.Finally,the gesture instruction recognition method based on BP neural network.During the experiment,the network was learned and trained and gesture instruction recognition was performed.On the basis of the original training,a second learning was carried out;and the momentum term was added when the weights of each layer were revised;experimental comparisons were also carried out to finally determine the most appropriate number of hidden layer nodes.Thereby improving the recognition efficiency,the average recognition rate is about 93.2%.