Two-dimensional motion detection model with parallel and layered multi-velocity channel

A new two-step 2-D motion detection model is presented. A 1-D multi-velocity motion detection layer is constructed based on the 1-D velocity detection feature of MIVD. A multi-velocity 2-D motion perception can be achieved by means of the good construction feature of cellular neural networks (CNN) to emulate the local dynamic interactions of cortex function columns. The experiment shows that this parallel and layered multi-velocity model is good at early vision motion perception. By using the model, the attention focus mechanism based on motion information drive can be realized, and the model can also be used as a part of objective-drive mechanism in high level feedback of vision system.