| Human-computer interaction, the core technology that enables a wide variety of applications such as smart home systems, somatosensory games, security and intrusion detection, has been becoming an indispensable part of people’s lives. Its interactive way has drawn public attention by intuitively, simply and naturally transferring people’s operation information to diverse equipment, and even become the focus of the academic and business. Nowadays, device-free gesture recognition, which doesn’t require people to carry any sensors or other devices, has been the current development tendency because of its simple operations and meeting people’s habits. Meanwhile, the existing Wi-Fi-based device-free gesture recognition methods, the hot of the research area, have fundamental limitations that make them failed in wide applications. Specifically, they utilize additional dedicated equipment, which are highly expensive. In this paper, we put forward a low-cost Wi-Fi-based gesture recognition system by using the commercial equipment. The specific research contents are as follows:(1) Signal preprocessing. Wi-Fi signal is susceptible to multipath propagation, signal attenuation and other interference factors the acquired data contains a lot of noise gesture. This paper adopts three steps to process the data that frequency domain fading compensation, filter out high frequency noise and smooth processing. Realized in filtering noise while the gesture data intact.(2) We propose an endpoint detection method based on Wi-Fi. In order to get fine-grained gestures to detect, we use a sliding window to process the signals to get gestures signals while smoothing the signals in stationary environments. Then, we use the threshold value to figure out the beginning point and the end point of the gesture. Finally, we get the exact gestures signals from all signals by using constraints.(3) We propose a new gesture recognition algorithm K-DTW. We propose an algorithm combining the clustering algorithm and classification algorithm. The algorithm can be devided into two stages. In the training stage, we use K-Means algorithm to cluster the characteristic matrix, and maps that to cluster the model signals. In the recognization stage:we calculate the distances between the features and the center of classes to minimize the matching scope and improve the accuracy.(4) Through the actual experiments we verify the feasibility and robustness of the algorithm. This paper compared the weak multipath interference and strong multipath interference scene to verify the robustness of the method. Different users with the same scene to verify the Ubiquitous of the system. The results show that the method can recognize the gestures well. |
