EchoSensor: Fine-Grained Ultrasonic Sensing for Smart Home Intrusion Detection

Abstract

This paper presents the design and implementation of a novel intrusion detection system, called EchoSensor, which leverages speakers and microphones in smart home devices to capture human gait patterns for individual identification. EchoSensor harnesses the speaker to send inaudible acoustic signals (around 20kHz) and utilizes the microphone to capture the reflected signals. As the reflected signals have unique variations in the Doppler shift respective to the gaits of different people, EchoSensor is able to profile human gait patterns from the generated spectrograms. To mine the gait information, we first propose a two-stage interference cancellation scheme to remove the background noise and environmental interference, followed by a new method to detect the starting point of walking and estimate the gait cycle time. We then perform the fine-grained analysis of the spectrograms to extract a series of features. In the end, machine learning is employed to construct an identifier for individual recognition. We implement the EchoSensor system and deploy it under different household environments to conduct intrusion detection tasks. Extensive experimental results have demonstrated that EchoSensor can achieve the averaged Intruder Gait Detection Rate (IDR) and True Family Member Gait Detection Rate (TFR) of 92.7% and 91.9%, respectively.