Trajectory annotation & reconstruction research platform
A two-phase research platform for pedestrian trajectory annotation and reconstruction analysis — help teach machines how people actually move.
Trajectory Annotation
Place strategic knots on pedestrian trajectories to mark significant movement points.
Real-time Annotation
Annotate pedestrian trajectories with precision using our intuitive tools.
Pattern Analysis
Compare and analyze trajectory patterns across different scenarios.
Knot Placement
Place strategic knots to mark significant trajectory points.
Trajectory Reconstruction
Reconstruct trajectories by placing anchor points and drawing curves through them.
Anchor Point Selection
Select anchor points from Phase 1 annotations to guide your reconstruction.
Curve Drawing
Draw smooth curves through anchor points to reconstruct the trajectory path.
Trajectory Validation
Help validate if knot placements capture essential trajectory information.
Phase 2 Demo Coming Soon
Research Purpose
A two-phase approach to understanding trajectory perception and reconstruction.
The Challenge
Autonomous vehicle safety relies on testing against diverse pedestrian behaviors. Current methods miss subtle variations and lack benchmarks for evaluating trajectory quality.
Phase 1: Annotation
Gather human annotations by placing knots on trajectories, capturing how people naturally perceive significant movement points and patterns.
Phase 2: Reconstruction
Validate annotation quality by reconstructing trajectories from anchor points, revealing if knot placements capture essential trajectory information.
The Impact
Comparing reconstructed trajectories with originals helps improve annotation guidelines and simulation accuracy for autonomous vehicle safety.
Collaboration
This project is a joint collaboration between Bangladesh University of Engineering and Technology (BUET), California Polytechnic State University (Cal Poly), and the University of California, Santa Cruz (UCSC).
Dr. Golam Md Muktadir, PhD, University of California, Santa Cruz
Dr. A. B. M. Alim Al Islam, Professor, Bangladesh University of Engineering and Technology (BUET)
Dr. Fahim Khan, Assistant Professor, Computer Science and Software Engineering, California Polytechnic State University