An LSTM-Based Approach for Understanding Human Interactions Using Hybrid Feature Descriptors Over Depth Sensors.