Distributed Vision Graph Update in Mobile Vision Networks

Previous work in the research field of Smart Camera (SC) networks intensively focused on algorithms for obtaining vision graphs in distributed camera networks consisting of static camera nodes. Each edge of the vision graph represents two cameras imaging a sufficiently large part of the same environment. In contrast, we propose that future SC networks will consist of static SCs, supported by mobile image providers (e.g. embedded devices equipped with image sensors such as mobile phones). Due to their mobility, location changes (especially turnings changing the direction of view) are very likely. In contrast to position changes, which can be detected by using off-the-shelf GPS sensors, turnings can only be detected by high-cost magnetic compasses. In this paper, we present a lightweight distributed algorithm for estimating turning movements based on visual sensor data and updating the system's vision graph. Our algorithm uses the optical flow as input to compute the turnings. Evaluations on a mobile image provider, consisting of an off-the-shelf embedded device equipped with an IP cam, show that our algorithm is able to estimate the angle of rotation with an accuracy of 80% and performs well in terms of CPU and memory utilization.