Outlier Recognition in Laser Vision Based on Robust Regression

Abstract

Aiming at the outliers of a laser vision system in the process of bar profile measurement, an outlier identification method based on Robust Local Weighted Regression(RLWR) and 3σ Criterion was proposed. In this method, Robust Local Weighted Regression is used to smooth the data, and then profile outliers are identified from the residual between the smooth data and the measured data according to 3σ Criterion. This method was used to identify the isolated outliers in the process of laser profile measurement, and compared with the moving mean identification method. The influence of different window widths on the recognition effect was discussed, and the ability of this method to identify multiple isolated outliers was verified. Subsequently, the profile data before and after the outlier processing was fitted, and the fitting effect and precision were analyzed. The analysis results show that the fitted ellipse after outlier processing is more accurate and the fitting accuracy is improved greatly. The identification method based on Robust Local Weighted Regression can effectively identify outliers in profile data. It is thus an efficient and robust outlier identification algorithm, important for improving the accuracy of laser profile measurement and achieve accurate measurement of profile size.

Keywords

• laser vision
• outlier recognition
• robust local weighted regression
• 3σ criterion