AI

Semantic Multimodal Compression for Wearable sensing Systems

Abstract

Wearable sensing systems (WSS's) are emerging as an important class of distributed embedded systems in application domains ranging from medical to military. Such systems can be expensive and power hungry due to their multi sensor implementations that require constant use, yet by nature they demand low-cost and low-power implementations. Semantic multimodal compression (SMC) mitigates these metrics in terms of data size by leveraging the natural tendency of signals in many types of embedded sensing systems to be composed of phases. In our driving example of a medical shoe with an insole lined with pressure sensors, we find that the natural airborne, landing, and take-off segments have sharply different and repetitive properties. SMC models and compresses each segment independently, selecting the best compression scheme for each segment and thus reducing total transmission energy.