Distinctive Features Generation for Perception of Global Motion

Abstract

The optical flow contains information such as motion and structure of objects. Theoretical analysis demonstrated these information could be retrieved by decomposing the flow into the primitive components of translation, divergence, rotation, and deformation. The popular physiological model of detectors for primitive components uses units each of which detects only any primitive component. We propose an alternative model for detecting primitive components, using a population code in which each unit is tuned to various motions. In our model,‘the relative velocity-space quartet’is generated in order to analyze motion features. On these spaces, measuring the gradients of relative velocity among the dots, primitive components can be retrieved. This paper reports the results of the computer simulation and discusses whether our model is consistent with the physiological knowledge of visual cortex.