Human motion analysis and action recognition
Adem Karahoca, Murat Nurullahoglu · 2008
The goal of human activity recognition systems is to build a system that can automatically infer a range of predefined activities, such as running, handclapping, etc, from recorded video sequences. Such a computerized system would be of great use for a variety of applications ranging from video surveillance for security to human-machine interaction. Typically, pattern recognition is the key component of such a system, where the goal is to classify, or more specifically recognize the data, based on a priori knowledge or statistical information extracted from the patterns. In this study, we presented a comparison of several well-known pattern recognition techniques for a human activity recognition system. We used Motion History Images (MHI) to describe these activities in a qualitative way and computed Hu moments, a widely used and well-known feature set to describe 2D or 3D shape, for further processing. Several feature extraction and classification methods were compared using different combinations and the results were analyzed. These methods are Principle Component Analysis and Linear Discriminant Analysis. Then we tested features with Support Vector Machines and K Nearest Neighbours classifiers.