Drones & AI - What's all the buzz about ?
summary
Demo combines DJI Windows SDK video feed with Microsoft Custom Vision models, using ffmpeg to slice drone video and an edge board (Jetson Nano/TX2) to run inference. Training needs hundreds of frames from the drone view; tested model detects koalas, crocodiles and mislabels wombats sometimes. Real numbers: IoT sensors ~70k/day, cameras ~28MB/day, drone imagery ~5GB and video ~10GB/day. Solution works end-to-end but is sensitive to lighting, battery life and current legal limits on beyond-visual-line-of-sight.
AI-written summary · about
chapters (4)
related talks
53mVideo Analytics for mere mortals with NVIDIA DeepStream SDK
45mArtificial intelligence: How data is evolving the future of technology
28mVision, Big Data and AI
01h 02mAI on the Edge
48m(Deep) Learning to Fly
20mComputer Vision at the Edge with OpenVINO
36mDeep Learning - Segmentation, Identification and Classification
22mThe media processing pipelines behind AI