This diagram from Zanker 2004, illustrates how the elementary motion detector, which detects motion in one dimension can be used to analyze two-dimensional scenes such a sequence of images or video.
Each pixel is an input for a detector in each direction, vertical and horizontal. The output from each dimension is calculated as vector components to determine the magnitude and direction of motion at each point.
When multiple frames are used, the delay is not a simple delay of a certain number of frames, but a temporal filter with a specified time constant that causes each current pixel to be some combination of the current and past frames, creating a smearing and delaying effect, when viewed visually.
