A neural implementation for high speed processing in linear stereo vision

The neural method for achieving real-time obstacle detection in front of a car using linear stereo vision is presented. The key problem is the linear stereo correspondence problem which consists in identifying features in two images that are projections of the same physical entity in the 3D world. The edge points extracted from each image are first classified into two classes. The problem is then decomposed into two identical sub-problems, each of them consisting in matching features of the same class. Each sub-problem is formulated as an optimisation task where an energy function, which represents the constraints on the solution, is to be minimised. Finally, a Hopfield neural network is used to solve the optimisation task. The preliminary classification of the edges allows one to implement the matching process as two networks running in parallel. Experimental results, using real stereo images, are presented to demonstrate the effectiveness of the proposed method.