Action recognition for a particular class category

[bit-scientist]

Let's say I want to have a fall detection solution that recognizes the action by skeleton information. Since that program doesn't consider any other class actions, is it better to train the model, say PoseC3D, with one class category? Other similar actions like sitting or standing (from sitting position) are in the class categories as well, should I also take those video samples too and train the network with 3 class actions (falling, standing and sitting)?
Thanks.