Computational Vision Model (CVM) Research and Development

In the present effort, we have built a new and improved model for the prediction of target acquisition. It is based on the latest models of early vision processes in the human visual system. It is a multi-channel model, based on three color opponent channels, and two temporal channels. Each channel is further subdivided into a set of multi-resolution channels. Computations are based on contrast ratio. Local energy values for these dimensionless contrast images are computed, and normalized by the sum of eye noise and clutter. The detectability metric is then computed from these energies. This model has been applied to the problem of detecting cars approaching intersections, and to the detection of mobile ground targets in imagery taken from an airborne first generation FLIR system. Reasonably good correlations were obtained. Several alternative studies were also undertaken, along with a critique, in hindsight, of the ideas and philosophy used in the development of the model.