Research Comprehensive Final Exam: Action Recognition on a Mobile Robotic Platform

We present an approach for recognizing human action from a mobile platform. Three components are essential to our action recognition pipeline. First, a new off-the-shelf active sensor is employed to obtain observations. Next, we use our observables to generate a set of pose-inspired features. Finally, we use these features to obtain a history over actions, with the use of a Hidden Markov Model and a Gaussian likelihood function. The proposed approach is insensitive to the changing lighting conditions which can be troublesome for color-based methods. To our knowledge, we are the first to develop a mobile, uninstrumented, color-invariant 3D bodymodel based approach for action recognition.