Processing streaming video for continuous monitoring
summary
How to build a robust continuous monitoring system that processes live-streaming camera video: handle stream faults (corrupt frames, interruptions), achieve real-time image-processing speed for alerts, and store processing results for later analysis. The talk demonstrates implementing such a streaming-capable pipeline, dockerizing the components, and creating a development pipeline for continuous monitoring (object detection, tracking, face ID use cases are mentioned as common examples).
AI-written summary · about
chapters (14)
- 0:00Introduction to project
- 3:19Final system results
- 8:54Architecture and workflow
- 10:03Continuous monitoring explained
- 10:46Object recognition with AI
- 19:38OpenVINO and pretrained models
- 21:02Docker in the project
- 23:08Video streaming processing
- 27:4524/7 stability challenges
- 30:34System demo
- 31:17Project resources and code
- 38:58Optimization and common errors
- 43:31Alerts and log analysis
- 49:15Q&A
related talks
53mVideo Analytics for mere mortals with NVIDIA DeepStream SDK
38mHow to Build Video Analytics for Live Streams Without Big Tech Resources
21mLeveraging Python for Real-Time Image processing
30mWhen performance matters: running a real-time computer vision pipeline
04mInterview
53mVideo Object Detection at Scale
34mTesting Live Video Streams At Scale
51mVideo Analytics for Football games