Hand's 3D movement detection with one handheld camera

This paper presents a scheme to create a real-time and reliable method for recognizing vision-based hand's 3D movement and to use the movement parameters for controlling 3D objects. The algorithm for 3D movement detection is totally based on analyzing feature points from the only camera in user's hand. As the algorithm is based on frames captured from one camera in untrained environment, it's difficult to distinguish similar movements on optical flow images, especially between shifting and rotating. A novel differentiation algorithm by voting from some weak classifiers is used. The algorithm provides a method of direct mapping user's hand movement to object control. We design an application of controlling a virtual 3D cube's movement and estimate the accuracy of the algorithm. And the experients' result presents that the 3D movement detection algorithm is efficient and robust enough for real-time interaction.