Method and system for activity classification

A method of identifying a most probable action performed by a person, the method comprising:

• transforming a dataset that includes two-dimensional (2D) positions of skeletal joints of the person by
• adapting the 2D positions into coordinates that are relative to a position of the person,
• scaling the 2D positions to be relative to a height of the person, and computing a velocity of each of the 2D positions;
• applying, to the transformed dataset, a first machine learning model that produces a first output that is representative of a most probable gesture identified from among a plurality of gestures;
• and
• applying, to the first output, a second machine learning model that produces a second output that is representative of the most probable action identified from among a plurality of actions;
• wherein the first and second machine learning models are neural networks that comprise at least one pair of inner product layer and rectified linear unit layer, followed by at least one long-short term memory layer, followed by a final inner product layer.

A computing device comprising:

• a camera that is configured to generate a video of a person;
• a processor;
• and
• a memory with instructions stored therein, that when executed by the processor, cause the computing device to
• generate, based on the video, a dataset that includes two-dimensional (2D) positions of skeletal joints of the person,
• transform the dataset by (i) adapting each of the 2D positions to be relative to a position of the person and (ii) scaling each of the 2D positions to be relative to a height of the person,
• apply, to the transformed dataset, a first neural network that produces a first output that is representative of a most probable gesture identified from among a plurality of gestures, and
• apply, to the first output, a second neural network that produces a second output that is representative of a most probable action identified from among a plurality of actions,
• wherein the first and second neural networks comprise at least one pair of inner product layer and rectified linear unit layer, followed by at least one long-short term memory layer, followed by a final inner product layer.