Positioning 3-D Objects by Using Gaussian Machines

Abstract

A simple stereo algorithm is presented to obtain the 3-D position of the object points in a scene. The algorithm can obtain the 3D position of an object point by merging a Gaussian machines (GM) neural model with the stereo vision system. Gaussian machines have graded output responses and stochastic behavior caused by random noise added to the input of each neuron so that the system can escape from the local minimum. In Gaussian machines the output function of a neuron is deterministic as in the Hopfield neural model, but the output value is influenced by random noise added to each input, thereby forming a probabilistic distribution. The key step in stereo vision is the correspondence process. The correspondence problem includes identifying features in two images that are projections of the same entity on the object. In this study a combinatorial optimization approach is used to resolve the correspondence problem for a set of features extracted from a pair of stereo vision systems. An energy function is defined to represent the constraints on the solution, which is then mapped onto a 2-D Gaussian machine neural network. Each neuron in the network represents a possible match between a feature in the left image and one in the right image. The network is said to be at its stable state when no change occurs in the states of neurons. Finally, the threshold of the neighborís disparity (NDT) is used to improve the precision of the correspondence situation. Once all the correspondence points are found, then obtaining the 3D object position is a simple matter of triangulation.