Reproduce “One Homography is All You Need: IMM-based Joint Homography and Multiple Object State Estimation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Paul Johannes Claasen, Johan Pieter de Villiers
Researchers at the University of Pretoria developed IMM-JHSE, an online Multi-Object Tracking algorithm that leverages a single initial homography estimate to enhance tracking robustness and affordability. The method jointly estimates target ground plane states and homography dynamics, achieving improved accuracy on datasets like DanceTrack and KITTI-car by dynamically adapting to camera motion and out-of-ground-plane object movement.
Key insights
- Incorporating a single initial homography estimate into a joint state vector, dynamically updated, can effectively bridge 2D and 3D MOT benefits, outperforming purely 2D methods without full 3D data requirements.
- Decoupling camera motion from target motion by directly including the homography in the track state and dynamically estimating it resolves a key modeling limitation, leading to more accurate ground plane tracking.
- An "IMM-like" adaptive association strategy, blending ground-plane and image-plane information, provides robustness to targets temporarily leaving the ground plane, maintaining track identities in dynamic scenes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.