Orientation learning for estimation of human pose direction from static image (パターン認識・メディア理解)

概要

論文の詳細を見る
The capability of estimating the human pose direction would be useful in many applications such as driver assistance system. In this paper, we consider the machine vision task of human pose direction estimation from static images based on learning the correct classification of a still image using adaboost algorithms. In our approach, we decided to classify the pose direction into one of 8 directions every 45 degree. To achieve our goal, we trained. 8 individual classifiers, each corresponding to one of the directions. We could assign human pose direction to one of the trained directions using these 8 classifiers. To evaluate the proposed method in its robustness and accuracy, the experiments have been performed to estimate the human pose direction between numbers of images which is highly challenging uncontrolled conditions in real world.
社団法人電子情報通信学会の論文
2010-03-08