Intelligence based error detection and classification for 3D measurement systems

For many years 2D machine vision has been used to perform automated inspection and measuring in the manufacturing environment. A strong drive to automate manufacturing has meant improvements in robotics and sensor technologies. So has machine vision seen a steady movement away from 2D and towards 3D. It is necessary to research and develop software that can use these new 3D sensing equipment in novel and useful ways. One task that is particularly useful, for a variety of situations is object recognition. It was hypothesised that it should be possible to train artificial neural networks to recognise 3D objects. For this purpose a 3D laser scanner was developed. This scanner and its software was developed and tested first in a virtual environment and what was learned there was then used to implemented an actual scanner. This scanner served the purpose of verifying what was done in the virtual environment. Neural networks of different sized were trained to establish whether they are a feasible classifier for the task of object recognition. Testing showed that, with the correct preprocessing, it is possible to perform 3D object recognition on simple geometric shapes by means of artificial neural networks