The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Spatial Analysis for Real-Time Video Processing with Adina Trufinescu
Oct 8, 2020
Adina Trufinescu, a Principal Program Manager at Microsoft, shares her expertise in computer vision and spatial analysis. She highlights groundbreaking innovations in real-time video processing that aid in monitoring movements and ensuring social distancing. The discussion touches on responsible AI practices to prevent misuse and the complexities of productizing advanced object detection technologies. Adina also delves into the iterative development of people counting models and the significance of evaluation metrics in enhancing performance accuracy.
AI Snips
Chapters
Transcript
Episode notes
Spatial Analysis Innovation
- Spatial analysis innovates by using computer vision AI on video streams, not just images.
- This allows for real-time analysis of people movement in physical spaces, like counting people and measuring social distancing.
Focus on People Movement
- The initial focus is on people movement due to customer demand in retail, manufacturing, and real estate.
- The system is designed to be flexible, allowing for future expansion to other object detection domains like vehicle analysis.
Recommended Hardware
- For large-scale deployments with many cameras, use Azure Stack Edge, a more robust device.
- The device prioritizes on-premises video processing to address privacy and data control concerns.