Automatic Object Recognition Using Combinational Neural Networks in Surveillance Networks

摘要

The level of recognition in humans is very high with less effort even for a multitude of objects in images, despite the fact that the image of the objects may vary with respect to view angle, size, translation and rotation. In this paper, a new scheme called combination neural networks is proposed which uses parallelism between two units in the recognition systems in application to visual sensor (surveillance) networks. The introduction of layered structure is a novel idea based on the concept of cache memory search of the CPU architecture. Objects can even be recognized when they are partially obstructed from views of visual sensors, i.e. partial occlusion due to the network topology of back propagation.