Event-Based Arm Pose Estimation
A pipeline for estimating arm joints from event-camera point clouds.
PythonPointNetMediaPipe
Context
This project explores arm pose estimation from event-camera data using the DHP19 and MoCA datasets.
Approach
The pipeline generates pseudo-labels from source videos with MediaPipe and trains a modified PointNet model to regress joint coordinates from event point clouds. It includes training and inference visualization scripts.
Scope and limitations
The datasets are not included in the repository, and the generated labels are estimates rather than independent ground truth. The repository describes the pipeline but does not report a validated accuracy figure.