Challenges of processing high-volume camera data streams
summary
Presentation describes the data-processing pipeline for the Spectro robot’s camera scanning system, focusing on obtaining synchronized frames from multiple cameras and how YDB was used to support that. It covers profiling tools that improved performance and explains why ROS2 was rejected for implementing the scanning system.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 2:15Manual inventory problem
- 4:44Robot Spectra
- 8:17Processing pipeline
- 11:47Actor-based architecture
- 14:41Hardware bottlenecks
- 17:32GPU workload optimization
- 20:24Model fusion efficiency
- 21:48Key lessons
- 24:22Future scalability
- 39:08System architecture
- 45:22Q&A
related talks
46mHow the Spectro robot scanning system works
24mBeyond the Frame: Using Computer Vision to Unlock the Potential of Video Data
21mLeveraging Python for Real-Time Image processing
17mData Challenges with 3D Computer Vision #bbuzz
53mProcessing streaming video for continuous monitoring
28mVision, Big Data and AI
47mHow we made Zivid Two - our 2nd generation 3D machine vision camera for robotics
27mAnalysis of sports scenes using computer vision models