Research On Non-Contact Measurement Method Based On Image Super-Resolution Reconstruction

In the factory production,in order to ensure the quality of the produced parts,a large quantities tests and inspections will be performed on the parts,mainly for the size measurement of the parts and the positioning of the components such as screws,pins,pins on the surface of the parts.Traditional detection technology solutions mainly use calipers,gauges and other tools.At the same time,manual observation and judgment are required.This method has low detection efficiency,and it is easily affected by fatigue and manual discrimination errors,resulting in low measurement accuracy.With the development of computer vision technology,computer visionbased part inspection technology has obvious advantages over traditional detection methods.Non-contact measurement of parts by taking part images,it can avoid problems such as surface scratches and deformation of parts.Such non-contact measurement method has the advantages of durable and stable detection.Therefore,the non-contact measurement method has gradually become the research focus of part measurement.Based on the positioning of pin grooves on the surface of a type of part,this thesis studies a non-contact measurement scheme.By using super-resolution reconstruction technology,the resolution of part images is increased,and the part measurement accuracy is improved.Network models such as SRCNN(Convolutional Network for Image Super-Resolution)and VDSR(Very Deep Convolutional Networks for Image Super-Resolution)are studied.By comparing parameters and running time,the VDSR model is selected as the Model suitable for non-contact measurement schemes.The VDSR model is optimized,the structure of the model is redesigned by increasing the model width,the processing layer of the original low-resolution image is added.The new layer is calculated in parallel with the original network.The optimized model can improve the subjective visual effect and peak signal-to-noise ratio of the reconstructed image.In order to make the scheme more widely applicable,the non-contact measurement scheme is extended.Due to the different parts sizes and different accuracy requirements,it is not possible to use a uniform specification inspection scheme.Therefore,it is necessary to design the part image acquisition process,which can be adapted to the local acquisition and stitching of parts of different sizes and improve the resolution of the part image.According to the stitching speed and stitching effects,feature point detectors such as SIFT(Scale-Invariant Feature Transform)and SURF(Speeded-Up Robust Features)are analyzed and compared.The SURF feature point detectors are selected as a suitable model for the non-contact measurement extension scheme.Design a non-contact measuring instrument model structure and design the measurement steps.The measurement scheme studied in this thesis can be used as a scheme for non-contact measurement of parts,which has realization value and good engineering application prospect.