Core-sets for Real-Time Tracking using Caratheodory Theorem, with Applications to Drones
summary
Proposes concrete, implementable coresets that compress point clouds so online pose-fitting and k‑segment trajectory compression run in milliseconds: a Carathéodory-based d+1 coreset (exact mean/combinations) and a practical sensitivity sampling scheme giving O(vcdim/ε^2) sample bounds. Presents a deterministic merge-and-reduce streaming tree with O(log N) coreset memory, real‑time 30Hz operation on Raspberry Pi + PS3 cameras, and experiments that match million‑dollar motion‑capture accuracy on quadcopters while avoiding randomized failure risk.
AI-written summary · about
chapters (10)
- 0:00Introduction to coresets
- 2:52Defining the coreset model
- 6:54Coreset construction theory
- 12:04Uniform vs. non-uniform sampling
- 15:14Mathematical foundations
- 20:14Real-world applications
- 24:18Streaming and cloud algorithms
- 29:36Robotics and pose estimation
- 37:23Technical implementation
- 43:41Future research and conclusion
related talks
10mEnabling Robot-assisted Motion Capture with Human Scale Tracking Optimization
21mVisualizing robot minds, and other fun stuff
24m3D Backscatter Localization for Fine-Grained Robotics
26mFrom Panoramic Vision to Lightweight 3D Worlds for Semantic Visual Tracking
19mAirways: Optimization-Based Planning of Quadrotor Trajectories according to High-Level User Goals
18mGridDrones: A Self-Levitating Physical Voxel Lattice for Interactive 3D Surface Deformations
48m(Deep) Learning to Fly
57mThe Convergence of Robotics, the Web, and IoT