SkyQuery: An Aerial Drone Video Sensing Platform
summary
SkyQuery is a complete aerial-drone video sensing platform: a domain-specific query language plus cloud video pipeline that runs on eight flights (three hours of video) and produces detection, tracking, merging, and gridded counts. It uses frame alignment with GPS+compass and optical flow to convert pixels to lat/long cheaply, a change-aware object detection method that finds small moving objects by comparing frames, and a forecast-based routing that prioritizes high-variance cells; alignment outperforms baselines and routing cuts error with fewer drones, parking-spot detection is surprisingly accurate.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:16Challenges
- 2:48Optimization
- 4:31Web Interface
- 4:41Query Language
- 5:12Data Types
- 6:50Parking Monitoring Example
- 10:40SkyQuery Techniques
- 14:18Conclusion
related talks
48mFlying services with a drone
13mFlyMeThrough: Human-AI Collaborative 3D Indoor Mapping with Commodity Drones
52mFlying services with a drone
28mGame of Drones
30mUAV-based Person Detection and Geolocation for Search and Rescue
28mVision, Big Data and AI
44mThe New Mash Up Platform
36mDeep Learning - Segmentation, Identification and Classification