【摘要】 基于深度图像可以方便的区分前景和背景,有效提高自然人机交互的性能。要实现面向大众的自然人机交互,需要深度摄像头在实时性、价格、适应环境等方面达到实用性。基于面结构光技术的深度摄像头实用性强,得到了迅速发展,其深度信息获取基于图像块匹配算法,在计算每个像素点的深度时,需要在测量范围内进行逐点搜索、图像块匹配、寻优,带来大量运算,这些运算要通过高性能计算机或专用并行运算芯片才能达到实时性,导致了深度传感器成本的增加。本文提出了一种全新的人体动作识别方法,即利用结构光图像直接进行动作识别,从而避免了复杂的深度图计算环节,降低了计算量,并能够充分利用图像的3D信息。该方法通过分析结构光图案的偏移所产生断裂点信息,成功实现了人手定位及人手动作识别,并将其应用于UI界面模拟鼠标操作。实验结果表明了算法的有效性和可行性。更多还原

【Abstract】 It is convenient to separate the foreground and background of the scene according to depth images,and thus can improves the performance of nature human- computer interaction( HCI) greatly. The popular use of HCI depends on the performance of the depth sensor such as price and real- time. The depth sensor based on surface structured light has high practicality and is applied widely. The depth information is gained using the matching algorithm of imgae block,which needs to search the optimal matching block in the image. In order to improve the efficiency of matching process,we need high- performance computer or special chip to achieve real- time performance. In this condition,the cost of the depth sensor is increased consequently. The paper proposes a new human action recognition method,namely directly action recognition based on structured light image. The algorithm not only reduces the amount of computation,avoids the complex depth map calculation,but also takes full advantage of the 3D information of the image. Through the analysis of breaking points,produced by the offset of structured light pattern,the method successfully positions human hands,detects gesture,and applies it to the UI interface,emulating mouse actions. Experimental results show the effectiveness and feasibility of the algorithm.更多还原