Body-part Detection for markerless 3D interaction game
(0) This paper proposes a novel human body part detection method based on statistical learning approach for marker-less 3D interaction game, which is capable of processing images extremely rapidly and achieving high detection rates.There are two key contributions.The first is the introduction of a new image feature called ”Omni-direction Feature”, which is a useful feature for objects that change direction freely, such as hand.The second is a parallel boosting algorithm based on AdaBoost.The parallel boosting algorithm will select a small number of critical visual features from a large set and yield extremely efficient classifiers.In this paper, an approach combining increasingly complex classifiers in a ”cascade” which allows non-body-part regions of the image to be quickly rejected while spending more computation on promising body-part-like area ”1” ”2” is used for parallel boosting.Experiment results demonstrate that using the Omni-direction features with the parallel boosting algorithm can achieve more rapid and robust body part detection both for training and testing than general Harr-like features based detection algorithm.
Harr Omni-direction Parallel Boosting
Xiangsheng Huang Xiubin Zhuang Jie Wang
Institute of Automation, Chinese Academy of Sciences, 100190 State Key Laboratory of Networking and Switching Technology, Beijing University of Posts and Telecom Control Science and Engineering, Hebei University of Technology, 300401
国内会议
长春
英文
235-242
2011-09-25（万方平台首次上网日期，不代表论文的发表时间）