A COMPARISON OF STATISTICAL REGRESSION AND NEURAL-NETWORK METHODS IN MODELING MEASUREMENT ERRORS FOR COMPUTER VISION INSPECTION SYSTEMS

Abstract

This paper compares measurement error models for computer vision inspection systems based on the statistical regression method and a neural network-based method. Experimental results demonstrate that both of the models can effectively correct the dimensional measurements of geometric features on a part profile. It also shows that the statistical regression method can perform excellent tasks when the functions for models are carefully selected through statistical testing procedures. On the other hand, varieties of neural network architectures all have good performance when training data are collected carefully. The explicit nonlinear relationship in neural network architectures is very effective in building a general mapping model without specifying the functional forms in advance. While statistical regression methods will continue to play important roles in model building tasks, the neural network-based method will be a very powerful alternative for precision measurement using computer vision systems.