Abstract: It is necessary to correct the nonlinear distortions in images of the machine vision measurement system which are vital to measuring precision. The method adopting a time-unvarying parameter model to correct the image distortions is difficult to achieve precise workpiece dimensions due to the stochastic variation of the distortions in production process. So a new nonlinear distortion correction method is proposed. The target part edge data series is described by a general expression for the linear and nonlinear auto-regressive time series model (GNAR model) and with the least square method and the minimum loss function criteria integrated with the computational complexity, the parameter estimation and order determination is realized. The distortion-free edges are obtained after model filtering and the workpiece dimensions are calculated through the vertical distances of the distortion-free edges. Experimental results show that the algorithm can be applied to machine vision measurement system for on-line measurement in machining process.

Key words: Distortion correction, Machine vision measurement, Time series model