摘要
In this paper we perform an applied comparative study of popular HOG based human detection and a state-of-the-art pose adaptive method that uses shape-based model construction. Both methods are implemented with kernel SVM, instead of linear SVM. Detailed performance evaluation is carried out on MIT pedestrian dataset and INRIA person dataset. This study shows that, although pose adaptive method has no significant advantage compared to the HOG based approach on those datasets, the pose adaptive approach is more efficient in detection and it has the capability to segment the human shape from images while carrying out detection which can be advantageous in many applications.
- 出版日期2012