Interpretation of Optical Flow through Neural Network Learning

In computer vision, the interpretation of optical neural network, it is assumed that motion develops all over the flow ( motion vector field calculated from images ) and estimation frame, centering around the center of the frame. of motion are important tasks. This study proposes a motion interpretation network which enables optical flow (OF) interpretation and describes motions on a plane through the use of a neural / r -. '. network with complex back propagation learning. Furthermore, r , ' \ (d.r,dy) an OF normalization network for optical flow normalization is t ! ' I proposed for the interpretation of diverse flow patterns, such as \ . ' d (d2 ,w) real image optical flow. Using test patterns and real image optical \ . d L flow, the generalization capacity of proposed network is investigated. And the ability is confirmed experimentally. Figure 1 Motion interpretation network