Human Activity Recognition using Deep Learning on Graphically Structured Data
You will have to install keras, tensorflow and all the necessary CUDA based drivers. Any of the latest combinations would do. Apart from these, you will have to install OpenPose, which is an open-source library for skeletal data extraction from the Videos. MATLAB is required to create the TAG from raw tracking data.
The code is briefly commented for now and we plan to elaborate their applications soon in a better structured manner.
TAG.pdf contains the thesis paper for Temporal Activity Graphs.
THESIS_REPORT.pdf contains our bachelor thesis project.
TAG_CODE directory has the necessary code to create a TAG from tracking data.
Tracking data is available on the respective dataset websites as mentioned in our bachelor thesis project.