A Sensor with Biological Preprocessing Features

The first part of the report reviews the basic geometric sensor configuration and its mapping to computation plane. Three basic configurations are discussed and their application to edge and short range motion detection follows. The next section covers a qualitative long range motion detection algorithm. The algorithm was tested first with synthetic images and next with real images, and in both cases the results were very satisfactory. The algorithm is now being tested multiple images. A one line correlation tracker is discussed in the next section. The correlation tracker can be implemented with parallel architecture and is, thus, very fast. tests have been performed with satisfactory results using real images. Finally pattern recognition using the BVS together with a class of transforms which are invariant to translation is discussed. It should be possible to implement this pattern recognition system using neural networks. The use of learning neural nets would allow recognition of similar but slightly distorted objects.