Sensor-based smart recognition system For wearable devices

Abstract

With recent advances in wireless sensor networks and embedded computing technologies, on body wearable devices such as smartphones and smart watches have become increasingly popular and play significant roles in our daily lives. The prevalence of smart wearable devices has sparked a new set of mobile computing applications that leverage the abundant information from sensors. In this thesis, I address three problems for wearab le devices. These three problems correspond to three different recognition tasks: to help the user recognize a subject (face recognition), to assist the device to authenticate the identity of another device (device pairing) and to authenticate user (user a uthentication). The first problem is to implement a robust and efficient face recognition system on smart glass. The main challenge is the tension between the high computation requirements of accurate face recognition algorithms and the resource onstraint s of smart glasses. To address this challenge, I propose a robust and efficient sensor -assisted face recognition system on smart glasses by exploring the power of multimodal sensors including the camera and Inertial Measurement Unit (IMU) sensors. Extensiv e evaluation results show the proposed system improves recognition accuracy by up to 15% while achieving the same level of system overhead compared to the existing face recognition system (OpenCV algorithms) on smart glasses. The second problem is to generate a cryptographic key for legitimate devices so that devices on the same user’s body can be paired together. I propose an automatic key generation system for wearable devices based on the user’s unique gait. The intuition of the proposed key generation a pproach is that the devices on the same body experience similar motion signals that are produced by the unique walking pattern of the user. Therefore, the unique gait signal can be exploited as shared information to generate secret keys for all on -body devices. The evaluation results show that the proposed system can generate a common 128-bit key for two legitimate devices with 98.3% probability. The third problem is to develop a gait-based user authentication system by using Kinetic Energy Harvesting (KEH). The main feature of the proposed system is it utilizes the output voltage signal of the energy harvester to achieve gait recognition rather than the accelerometer. Compared with traditional accelerometer - based gait recognition system, the proposed system can reduce energy consumption by 78% while achieving comparable recognition accuracy.