<![CDATA[This paper proposes an approach for cal- culating and estimating human body orientation using geometric model. A novel framework integrating gradient shape and texture model of the human body orientation is proposed. The gradient is a natural way for describing the human shapes, while the texture explains the body characteristic. The framework is then combined with the random forest classifier to obtain a robust class differ- ence of the human body orientation. Experiments and comparison results are provided to show the advantages of our system over state-of-the-art. For both modeled and un-modeled gradient-texture features with random forest classifier, they achieve the highest accuracy on separating each human orientation class, respectively 56.9% and 67.3% for TUD-Stadtmitte dataset.]]>
Copyrights © 2015
