Histograms of Oriented Gradients is a well known and applied descriptor, however “black box” use is common. Gradient computation is the key to performance and may be application dependent. In this paper we examine explicit, implicit and Hessian schemes as opposed to the recommended centred mask. Results indicate the explicit Bickley scheme boosts robustness, both static and dynamic information are important to recognition and full body Gait-Energy Images are preferred. Robustness is boosted by specific choice of cell and bin parameters and SVM where actions are pre-classified using temporal information.
|Title of host publication||Advances in Visual Computing|
|Subtitle of host publication||9th International Symposium, ISVC 2013, Rethymnon, Crete, Greece, July 29-31, 2013. Proceedings, Part I|
|Editors||George Bebis , Richard Boyle, Bahram Parvin, Darko Koracin, Baoxin Li, Fatih Porikli, Victor Zordan , James Klosowski , Sabine Coquillart, Xun Luo , Min Chen , David Gotz|
|Number of pages||10|
|Publication status||Published - 2013|
|Name||Lecture Notes in Computer Science|