Lightning Talk: Streaming and Processing Edge Vision Data in Real Time
summary
Edge vision systems run lightweight models locally and send only events or metadata upstream, claiming up to ~25% revenue loss from empty shelves and solving it with YOLO on TensorFlow Lite for continuous local inference. Cheap sensors act as watchdogs to trigger just-in-time CV for cases like whale detection, reducing compute and false positives. A stadium queue app runs fully offline peer-to-peer and optionally syncs later. Network and API stack uses WebRTC and gRPC with Raspberry Pi + Coral TPU acceleration.
AI-written summary · about
related talks
59mML and the IoT: Living on the Edge
14mComputer vision on edge devices
14mLive Streaming Object Detection by Leveraging GPU with AVA Platform on EWAOL
30mWhen performance matters: running a real-time computer vision pipeline
53mVideo Analytics for mere mortals with NVIDIA DeepStream SDK
25mWhats the Best Way To Reduce Storage Requirements Without Losing Insights? Push AI To...
46mTransparent Open Source AI Video Analytics with Panfrost
15mUsing The Power of Edge Compute To Enhance Streaming Delivery