【作者】 柴秀娟； 唐毅力； 山世光； 陈熙霖； 马丙鹏；

【Author】 Xiujuan Chai1,Yili Tang1,Shiguang Shan1,Xilin Chen1,Bingpeng Ma2 1 Key Laboratory of Intelligent Information Prosessing of Chinese Academy Sciences (CAS),Institute of Computing Technology,Beijing 100190,China 2School of Computer Science and Technology,Huazhong University of Science and Technology,Wuhan 430074,China

【机构】 中科院计算所中科院智能信息处理重点实验室； 华中科技大学计算机科学与技术学院；

【摘要】 基于视觉的手势分析与理解是实现新一代人机交互的关键技术,而复杂环境下的手势检测、跟踪等一直是手势分析实用化的瓶颈问题。为解决复杂背景条件下的手势跟踪问题,本文融合图像信息和深度信息,提出深度限制的图像表观特征建模方法,有效去除背景区域的影响,提取前景目标的紧致描述,在核跟踪的框架下,实现物体的鲁棒、快速跟踪。与传统的单独利用2D图像信息的跟踪方法相比,本文提出的多通道信息融合的策略可获取更为精确的跟踪精度。并且,本文设计并实现了一个具体的应用场景,来验证手势交互的可用性。更多还原

【Abstract】 Vision-based gesture analysis and interpretation is a key technology to realize the next-generation human computer interaction.Hand detection and tracking under complex enviroment has been the bottle-neck in practice of gesture analysis.To tackle the hand tracking problem under complex background,this paper explores to integrate the image and depth information.A novel depth-constraint image appearance featue modeling mehod is proposed to eliminate the background and extract more compact discription of foreground object.The robust and fast object tracking is implemented under the kernal-based tracking framework.Compared with traditional method by using only image information,the proposed multi-modal information fusion-based strategy obtains much higher tracking accuracy.Also,a concrete application scenario is designed and implemented to show the usability of gesture interaction.更多还原