Research And Implementation Of Flat Sheet Metal Image Mosaics

The size measurement of sheet metal parts based on computer vision has the advantages of high efficiency,low cost and automation.For large-size sheet metal parts exceeding the field of the imaging system's view,it is necessary to collect images with overlapping areas in segments.Then synthesize the whole sheet metal image by image mosaic.This paper focuses on image mosaic technology based on phase correlation method and point feature matching and deep learning.The main works are summarized as follows:(1)A robust contour-based phase correlation method is proposed for image mosaic of sheet metal parts with sparse features and symmetrical outline.Firstly,select equal-size areas containing overlapping area and extract their outlines.Secondly,registration parameters of the contour images are obtained by using phase correlation method.Finally,the registration parameters of contour images are transformed to ones of original sheet metal parts.The sheet metal part images are stitched according to the registration parameters.Experimental result shows that this stitching method has advantages of strong anti-interference ability and fast speed for sparse features and contour-symmetrical sheet metal images.Also accurate registration parameters and stitching effects are obtained.(2)Aiming at the poor results that many image mosaic methods based on the feature points are used to sheet metal part images with few features,an improved method is proposed.Firstly,the Fast feature points are extracted and screened.Secondly,the gray feature are sampled by using template region.Set the rotation angle and scaling ratio as search regions,calculate the structural similarity index measurement(SSIM)value and then accomplish the points matching.Finally,the registration parameters are obtained by the points matching results.And the 3?principle is used to remove abnormal values.The experiment results show that the accurate registration parameters and image mosaic effect can be obtained under the condition that the angle search domain is[-45~?,+45~?]and scaling search domain is[0.5,1.5].(3)Aiming at the poor results that many image mosaic methods based on the feature points are used to sheet metal part images with weak texture,few features and many similar regions,an improved method is proposed.Firstly,the Fast feature points are extracted and screened.Secondly,the gray feature is sampled by using template region.Complete feature point matching by using an unsupervised learning method based on the principles of GANs.Finally,the registration parameters are obtained by the points matching results.And the 3?principle is used to remove abnormal ones.The experiment results show that the accurate registration parameters and image mosaic effect can be obtained under the condition that the angle search domain is[-40~?,+40~?]and scaling search domain is[0.5,1.5].