Human Posture Detection in Always-on Body Worn Cameras

Abstract

Internet of Things (IoTs) is the emerging Technology as the third wave in the development of the internet. IOT is accepted to have massive impact sensors, surveillance devices; wearable's and body area networks with advanced interface technique. This leads to the need for a hardware-software co-design approach. We present a highly parameterized hardware design allowing configurability and the ability to evaluate different design choices in a short amount of time. This paper aims to analyze the use of Deep learning algorithm for building such reconfigurable low power devices. We develop the hardware architecture for such a system and provide experimental results obtained for the case study of Posture detection for body worn cameras used for law enforcement. Our proposed system, implemented on a Raspberry picamera platform uses minimal power consumption with an accuracy of more than 90% for the selected case study of Posture Detection. 