Automatic Inspection and Processing Based on Vision Stitching and Spectral Illumination

Abstract

The study investigates the automatic inspection and processing of bicycle components based on vision stitching and spectral illumination. Vision stitching mainly involves the use of algorithms of white balance, 
scale-invariant feature transform (SIFT) and roundness for automatic accessory inspection of the whole component image. The illumination intensities, angles, and spectral characteristics of light sources are analyzed using a spectrometer. Unrealistic color casts of feature inspection for global automatic adjustment are removed using a white balance algorithm. For the stitching of large images, SIFT is used to extract and detect the image features. The Hough transform is used to detect the roundness of the bicycle component. The inspected features include geometry size, roundness, and image stitching. Results showed maximum errors of 0°, 10°, 30°, and 50° for the spectral illumination of white light LED arrays with respective differential shift displacements of 4.4, 4.2, 6.8, and 3.5%. The deviation errors of image stitching for the stem accessory in x and y coordinates are 2 pixels. SIFT and random sample consensus (RANSAC) enable the transformation of the stem image into local feature coordinates.