Kitware will present the paper ‘Human Action Recognition in Large-Scale Datasets Using Histogram of Spatiotemporal Gradients’ by Kishore K. Reddy, Naresh Cuntoor, Amitha Perera, and Anthony Hoogs.
With Michael Ryoo from the Jet Propulsion Laboratory at CIT, Anthony Hoogs, Arslan Basharat and Sangmin Oh will present the tutorial ‘Activity Recognition for Visual Surveillance‘.
This tutorial provides an overview of human activity recognition approaches and their applications to surveillance systems. A scale-based taxonomy, discussing activity recognition at various scales from high-resolution videos of indoor environments to very low-resolution imagery from aerial platforms is presented. Various types of recognition methodologies for activities involving multiple humans and objects (e.g., vehicles) are introduced, along with corresponding target surveillance applications. Various state-of-the-art approaches in the activity recognition literature are covered. Another important component of the tutorial is a review of various existing public activity recognition datasets and several success stories of activity-aware surveillance systems in practice. Audience will be introduced to the limitations of some of the contrived datasets and challenges of some of the real-world datasets. These challenges include, but are not limited to, variations in fields-of-view, various types of occlusion, camera motion, low image resolution and frame rate.