Human posture recognition with the stochastic cognitive RAM network

Abstract

This paper examines a weightless neural network (WNN) for human posture recognition. Like all earlier weightless neural network models, the Cognitive RAM Network (CogRAM) learns in one pass through the data and due to its simplicity it can be fabricated in hardware. While it has shown good performance in earlier studies, it still suffers from the common problem of network saturation especially when it comes to high dimensional and poorly separated data in the feature space. Hence, we proposed the Stochastic CogRAM which has shown significant improvements when tested on the challenging human postures recognition problem. We also present some comparisons of the experimental results obtained from the popular K-Means clustering algorithm. Future research is outlined at the end of the paper. © 2012 Springer-Verlag.