Neural network computing for interpretation of novel sensor signals for six-degree-of-freedom motions of objects

Abstract

A sensor modelling via an artificial neural network is presented in this paper. The sensor is an optical type which is designed to measure absolute three-dimensional positions and orientations of objects in six degrees of freedom (DOFs), utilizing a triangular pyramidal mirror having an equilateral cross-sectional shape referred to as a three-facet mirror, a He-Ne laser source, and three position-sensitive detectors. We can get the 6-DOF motion of any object simply by mounting the three-facet mirror on the object; however, it takes rather a long time to determine the 6-DOF pose of objects in motion at any instant since the conventional method uses an iterative estimation algorithm. Due to this low calculation speed the previous method may not be effectively applied to real time applications. To overcome this limitation a multi-layer perceptron is constructed and trained for fast calculation in this paper. The calculation results of the original iterative method and the neural network model are compared with each other. From the comparison, the neural network model is proved to be sufficiently accurate and fast to be suitable for real time applications.