Research On Gesture Recognition Based On CW Distribution And Improved FOA

With the development of smart devices' hardware and software, the importance of human-computer interaction has gradually emerged. The interaction between people and the machine has a great impact on the efficiency of work. Researchers study the main purpose of human-computer interaction technology is to achieve efficient interaction between people and computers. And because of its flexible and efficient features, hand gesture has become the hotspot of human-computer interaction in recent years.Hand gesture recognition is mainly divided into three methods which is sensor-based, computer image based and wireless signal based. The performance of the first two methods has much higher requirement of the environment and equipment which is hard to fulfil. So in this paper we studied the method of hand gesture recognition based on WiFi signal.To capture and analyze WiFi signals, we need technologies about soft defined radio. We set up soft defined radio system based on SORA and PC,collected WiFi raw data using 802.11a protocol. After analyzing the physical layer frame structure of the WiFi signal, we selected long training sequence as the gesture data source for feature extraction. To extract the information which appeared both in time domain and frequency domain we used Choi-Williams distribution (C WD) which is one of time-frequency analysis to transform the signal into time-frequency domain. Then four feature vectors are extracted as support vector machine (SVM) input.There are two key parameters that has great influence on classification result, to obtain the most proper parameters in this scene, we used improved fruit fly algorithm (FOA) to optimize these parameters. Considering that the number of fruit fly individuals and the scope of the search are fixed throughout the iteration which may cause the decrease of convergence speed and optimization precision. We proposed an improved FOA based on dynamic group and direction correcting (DPDC-FOA) and tested this algorithm with 6 benchmark functions. The result shows DPDC-FOA has better performance in speed and accuracy of convergence. Then this improved FOA is applied to the SVM parameter optimization. The result shows that after parameter optimizing, all four feature vectors' recognition accuracy is improved. The highest average recognition rate has reached 96.8%. This result shows that gesture recognition algorithm based on Choi-Williams distribution and improved FOA proposed in this paper has certain practicability and research significance and provides a new idea for solving related problems.