Human Movement Analysis from the Egocentric Camera View
Mohammad Almasi · 2020 International Conference on Electronics and Sustainable Communication Systems (ICESC) · 2020
In this study, a method of human movement analysis is introduced according to making and utilizing human motion datasets in a first-person egocentric viewpoint using head-mounted cameras. Two steps of virtual and real tests are required to build the dataset. The blender has been used to capture the actions in the virtual mode with five thousand frames from various movements in different scenes to assure the reliability of the algorithm. Furthermore in real test 170000 frames captured in different actions from seventeen people. To recognize each act, the long short term memory architecture was utilized for the classification of the training database. In the process of this stage, the optical flow is utilized for obtaining the head and the concordance camera position, combined with the scene which is utilized for recognizing the motion pattern for each person.