SIGCOMM 2020 - Reducto - On-Camera Filtering for Resource-Efficient Real-Time Video Analytics

Explore cutting-edge techniques for resource-efficient real-time video analytics in this conference talk from SIGCOMM 2020. Dive into the innovative Reducto system, which employs on-camera filtering to optimize query pipelines and enhance performance. Learn about binary classifiers, pixel-level differences, and threshold-based approaches for efficient video processing. Discover the system design, implementation results, and potential applications of Reducto. Engage with discussions on updating hash tables, utilizing additional resources, handling moving cameras and frames, and leveraging heterogeneous computing. Compare Reducto's capabilities with other systems like DDS and gain insights into the future of video analytics technology.

Introduction
Trends in Video Analytics
Query Pipeline
Filtering
Key Question
Binary Classifiers
Pixel Level Differences
Reducto
Threshold
System Design
Results
Conclusion
Questions
Updating the Hash Table
Using More Resources
Moving Cameras
Moving Frames
Heterogeneous Computing
Reducto vs DDS
Final Question