Pose2Action

Action recognition using poses

code

The task here is to recognise the actions done by a human in an input 2-D video. The pipe line fist performs the 3-D pose estimation and later uses it as feature vectors to perform classification, the classifier is based on LSTM networks and experiments were performed to check whether using 2-D projections of poses instead of 3-D projections can still yeild valuable results.