Date: Wednesday, May 18 (Main Conference Day 2)
Start Time: 11:25 am
End Time: 11:55 am
Machine learning (ML)-based vision edge AI has wide applicability across a variety of segments, including consumer electronics, home security, smart buildings, smart city and factory automation. To date, most vision edge AI implementations have focused solely on vision–detecting people, objects and activities. Moreover, implementations have suffered from high power consumption, typically requiring AC power. These two factors have limited the penetration of vision edge AI. In this presentation, we describe a modern approach based around the Katana low-power edge AI SoC that improves performance and optimizes power consumption by fusing together inputs from a variety of sensors (including vision, sound and environmental, among others) into an AI processor running multiple ML models in parallel. We will show how this approach enables the design of more intelligent, context-aware, battery-powered edge AI inference devices, significantly broadening the usefulness and penetration of vision edge AI across multiple markets and new applications.