Wearable, energy-opportunistic vision sensing for walking speed estimation

State-of-the-art wearable systems are typically performance-constrained, battery-based devices which can, at most, reach self-sustainability using energy harvesting and aggressive duty-cycling. In this work, we present a wearable vision sensor node which can reliably execute computationally-intensive computer-vision algorithms in an energy-opportunistic fashion. By leveraging a burst-generation scheme, the proposed system can efficiently provide the energy guarantees required for tasks with temporal dependencies, even under highly variable harvesting conditions. By mounting the node on a user's glasses, the node is able to acquire a sequence of images and determine the user's walking speed, requiring only a small solar panel and capacitor. Both hardware and software have been fully optimized for ultra-low power consumption and high performance. Extensive experimental results show the energy node's energy proportionality and the accuracy of its walking speed estimation.