在下述基本假设下提出了一种基于单目相机结构的人手跟踪算法:(1)大拇指根始终可见;(2)除大拇指外,至少有另外一个手指是可见的。首先,利用UKF滤波技术。获取人手的全局运动,然后,利用3-D约束条件乖最小化代价函数获取各关节的局部运动。实验表明,算法可以部分地解决人手自遮挡问题,具有令人满意的跟踪精度。
A novel human hand tracking algorithm was put forward based on a single-view camera under the following two assumptions: (1) the root of the thumb is visible; (2) at least one of the five fingers is visible except the thumb. First, obtain the global movement using Unscented Kalman Filter from the visible fingers, identifying an affine transformation between adjacent frames, and then get the local movement using 3-D constraints and cost minimization proposed. The experiments show that the algorithm can deal with hand self-occlusion problem to some extent and has satisfactory tracking accuracy as well.