ANN IMPLEMENTATION OF STEREO VISION USING A MULTILAYER FEEDBACK ARCHITECTURE

An Artificial Neural Network (ANN), consisting of three interacting ne ural modules, is developed for stereo vision. The first module locates sharp intensity changes in each of the images. The edge detection pro cess is basically a bottom-up, one-to-one input-output mapping process with a network structure which is time-invariant. In the second modul e, a multilayered connectionist network is used to extract the feature s or primitives for disparity analysis (matching). A similarity measur e is defined and computed for each pair of primitive matches and is pa ssed to the third module. The third module solves the difficult corres pondence problem by mapping it into a constraint satisfaction problem. Intra- and inter-scanline constraints are used in order to restrict p ossible feature matches. The inter-scanline constraints are implemente d via interconnections of a three-dimensional neural network. The over all process is iterative. At the end of each network iteration, the ou tput of the third constraint satisfaction module feeds back updated in formation on matching pairs as well as their corresponding location in the left and right images to the input of the second module. This ite rative process continues until the output of the third module converge s to an stable state. Once the matching process is completed, the disp arity can be calculated, and camera calibration parameters can be used to find the three-dimensional location of object points. Results usin g this computational architecture are shown.