ABSTRACT
This chapter motivates the work towards a study on versatile recognition using Haar-like features for human sensing. It discusses several potential target applications. It also discusses on the characteristic constraints of a system comprising wireless sensor nodes. It mentions the significance of smart sensing. The chapter brings together new algorithms and insights to construct a versatile recognition framework that can be used to recognize patterns from sound, acceleration signals, and image at low calculation cost. It proposes a novel feature extraction method that allows to handle acceleration signals using Haar-like feature extraction framework. It investigates the classifier design for versatile recognition. A versatile recognition algorithm processes image, sound, and 3-Dimension acceleration signals with the common framework at low calculation cost. The chapter proposes 1-Dimension Haar-like feature to roughly estimate frequency information of the temporal signals. It also proposes biaxial and mean-embedded Haar-like features to extract standard deviation and interaxial correlation from 3-Dimension acceleration signals in Haar-like feature framework.
