Authors: Jiashuang Xu; Zhangjie Fu; Xingyue Du
Addresses: Computer and Software College, Nanjing University of Information Science and Technology, No. 219, Ningliu Road, Pukou District, Nanjing City, Jiangsu Province, China ' Computer and Software College, Nanjing University of Information Science and Technology, No. 219, Ningliu Road, Pukou District, Nanjing City, Jiangsu Province, China ' School of Humanities and Social Sciences, Xi'an Polytechnic University, No. 58, Shangu Avenue, Lintong District, Xi'an City, Shaanxi Province, China
Abstract: Currently, the contactless human-computer interaction (HCI) has become a heated research topic due to the springing up of the novel intelligent terminals. The existing interaction systems are used to adopt depth cameras, motion controller, radio frequency devices. The common drawback of the above approaches is that all the participants are required to obey the unistroke writing standard for data acquisition. Thus, we are motivated to propose a more adaptive, contactless graffiti-writing recognition system with channel state information (CSI) derived from Wi-Fi signals. We extract the unique CSI waveform caused by writing action to represent each letter. To cater to more users' writing customs, we train separate hidden Markov model (HMM) for eight of 26 letters and conduct cross-validation for testing. The average detection accuracy reaches 94.5%. The average recognition accuracy for the 26-letter model is 85.96% when the number of training samples is 100 from five subjects. The real-time recognition efficiency measured by characters per minute (CPM) is 11.97 (= 31/155.24 s).
Keywords: air-write recognition; wireless sensing; channel state information.
International Journal of Computational Science and Engineering, 2020 Vol.21 No.2, pp.163 - 172
Received: 18 Aug 2017
Accepted: 17 Dec 2017
Published online: 06 Mar 2020 *