Action in Mind: A Neural Network Approach to Action Recognition and Segmentation

Recognizing and categorizing human actions is an important task with\napplications in various fields such as human-robot interaction, video analysis,\nsurveillance, video retrieval, health care system and entertainment industry.\nThis thesis presents a novel computational approach for human action\nrecognition through different implementations of multi-layer architectures\nbased on artificial neural networks. Each system level development is designed\nto solve different aspects of the action recognition problem including online\nreal-time processing, action segmentation and the involvement of objects. The\nanalysis of the experimental results are illustrated and described in six\narticles. The proposed action recognition architecture of this thesis is\ncomposed of several processing layers including a preprocessing layer, an\nordered vector representation layer and three layers of neural networks. It\nutilizes self-organizing neural networks such as Kohonen feature maps and\ngrowing grids as the main neural network layers. Thus the architecture presents\na biological plausible approach with certain features such as topographic\norganization of the neurons, lateral interactions, semi-supervised learning and\nthe ability to represent high dimensional input space in lower dimensional\nmaps. For each level of development the system is trained with the input data\nconsisting of consecutive 3D body postures and tested with generalized input\ndata that the system has never met before. The experimental results of\ndifferent system level developments show that the system performs well with\nquite high accuracy for recognizing human actions.\n