Human Activity Recognition Using Sequences of Postures
Vili Kellokumpu, Matti Pietikäinen, Janne Heikkilä · 2005
This paper presents a system, which is able to recognize 15 different continuous human activities in real-time using a single stationary camera as input. The system can recognize activities such as raising or waving hand(s), sitting down and bending down. The recognition is based on describing activities as a continuous sequence of discrete postures, which are derived from affine invariant descriptors. Using affine invariant descriptors makes our system robust against such differences in camera locations as distance from the object and change in viewing direction as these differences can be considered to have the affect of near affine transformations as human silhouettes are considered. 1