Check out RSS, or use RSS reader to subscribe this item
Confirmation
Authentication email has already been sent, please check your email box: and activate it as soon as possible.
You can login to My Profile and manage your email alerts.
Sponsored by the Center for Science and Technology Development of the Ministry of Education
Supervised by Ministry of Education of the People's Republic of China
This paper studies the problem of robust classifier for visual pedestrian detection, based on HOG (Histograms of Oriented Gradient) descriptors. The average gradient image over the training positive examples is similar with the personal sketch and the variance represents poses in various complicated environments. Inspired by these, Large Margin Distribution Learning is investigated. It is found that the performance of LDM classifier with appropriate parameters is better than SVMs', nevertheless it is difficult to obtain optimal parameters. In order to improve robustness, ELDM (Ensemble Large Margin Distribution Machine) including maximizing the minimum margin and two margin distributions is proposed, which is the best in convex combination of SVMs and two LDMs with different parameter orders. The impact of parameters which are sensitive in the robustness of LDM is weakened by ELDM and ensemble parameters are obtained by the mesh grid optimization. Experimental results show that ELDM can obtain better results than existing classifiers for pedestrian detection in the classification performance and stability.