Pose Estimation

Performs pose estimation from depth images through body part classification and joint estimation.

We train TensorForests with depth signals to predict body parts. These predictions are then used to estimate 3D joint positions.

Instructions

1. Put pose batch files (batches of depth images) named as: Infant-batch-X (X = 0, 1, 2 ...) inside data/raw/train/

2. Check/Edit the configuration inside config.py

3. Run body_part_classifer_prototype.py. It trains random forests from the batch files and outputs joint predictions.

References

Shotton, Jamie, et al. "Efficient human pose estimation from single depth images." IEEE Transactions on Pattern Analysis and Machine Intelligence 35.12 (2013): 2821-2840.