Abstract
•Information can be represented by inhalation, exhalation and breath-holding.•A new signal processing method is proposed to realize perception and recognition of respiratory patterns.•Recursive or pipelining processing facilitates the implementation of real-time systems.
Patients with neuromuscular system disease may be systemic paralyzed and even lose the ability to speak. Patients can use breath to express intentions. But existing solutions either only allow patients to express several predefined sentences or cannot serve blind patients. We aim to develop a breath-based text input system named ExHIBit that allows paralyzed users, even paralyzed blind users, to freely express any sentences composed of letters and numbers. Based on the above motivation, we adopt binary sequences to express letters/numbers, and then specify that the inspiratory and expiratory actions represent one bit information respectively. Breath-holding action serves as an indication of symbol value change (0⇄1). We leverage commercial RFID devices to capture user’s respiratory waveform. A novel recursive algorithm called Dual-Channel Follow-Keep-Pump (DFKP) is proposed to extract channel signals of respiratory waveform. Then we adopt the first-order differential features of channel signals to detect and identify respiratory actions. ExHIBit adaptively adjusts parameters to cope with the non-stationary changes of respiration. Experiments demonstrate that ExHIBit can obtain 95.14% recognition accuracy and an input speed of 1.86 word/minute.