P-field: a computational model for binocular motion processing

In a binocular camera system moving through the environment, it is possible to obtain a three-dimensional field of vectors, where each vector is parallel, to the actual inducing 3-D motion of an imaged point (relative to the moving camera), and scaled in magnitude by the depth of that point. The authors derive this 3-D vector field (the p-field), discuss its computation from the image quantities of optic flow and disparity, and examine its behavior for specific cases of motion as well as for general motion. They present experimental results in computing this vector field and discuss issues related to the determination of the motion parameters from it. The results are demonstrated on synthetic and real data.<
>