LoCaF: Detecting Real-World States with Lousy Wireless Cameras

Abstract

The Internet of Things (IoT) integrates wireless sensors to provide online and real-time access to the state of things and places. However, many interesting real-world states are difficult to detect with traditional scalar sensors. Tiny wireless camera sensor nodes are an interesting alternative as a single camera can observe a large area in great detail. However, low image resolution, poor image quality, and low frame rates as well as varying lighting conditions in outdoor scenarios make the detection of real-world states using these lousy cameras a challenging problem. In this paper we introduce a framework that addresses this problem by providing an end-to-end solution that includes energy-efficient image capture, image enhancement to mitigate low picture quality, object detection with low frame rates, inference of high-level states, and publishing of these states on the IoT. The framework can be flexibly configured by end-users without programming skills and supports a variety of different applications.