Human Action Recognition based on Motion Velocity

In this article, we propose a model for human limbs detection and action recognition. The model simulates the functionality of the visual system to identify speeds and directions of any motion by using computational mechanisms underlying the perception of biological motion. Here, local motion and motion velocity of human limbs are investigated. Two different types of cells in the visual cortex are modeled and presented in two stages. The first stage simulates the functionality of the primary visual cortex where the response of this area drives from the local flow motion, while the second stage mimics the responses of the middle temporal cells where these cells respond strongly to the motion velocity. The model detects eight different directions and two speeds. Our model was tested and evaluated using different kinds of actions where good results are achieved.