Based on PyTorch library, realizing human activities recognition using 2D skeleton joint points
- Basic idea is similar with RNN-for-Human-Activity-Recognition-using-2D-Pose-Input: to classify human activities using a 2D pose time series dataset like skeleton joint points which can be detected by some software such as OpenPose.
- Realizing LSTM network by PyTorch
- Dataset is a subset of the Berkeley Multimodal Human Action Database (MHAD)
- Thanks to RNN-for-Human-Activity-Recognition-using-2D-Pose-Input, dataset can be downloaded from raw data
- Put the dataset into data/HAR_pose_activities/database/
- Detail is in lstm.ipynb
- Not bad result is realized in a 8 Minutes training (TitanXP GPU)