Study On Image Processing Methods In Multi-view Stereo Vision Measurement For Large Forgings

Large forging is the basis of heavy equipment manufacturing industry in our country, which reflects the comprehensive strength of national manufacture level. Large forging is usually used in key parts and components in engineering fields to ensure their physical and mechanical properties in extreme conditions. However, limited to the level of large hot forgings'manufacturing, which has low material utilization ratio and large energy consumption, the accurate and reliable measurement method for geometric parameters of the forging is needed to plan the forging process reasonably and improve the machining accurately.In complex forging workshop, it is difficult to obtain the precise and real-time dimensions of large hot forgings. In this paper, multi-view stereo visual system is presented, where two visual sensors constituted by binocular vision collect forging's feature from different view, and the feature emphasized by laser stripes. These features are transformed to a same coordinate system using registration technique, so that the whole dimensions of large hot forging can be acquired. This paper launches the research from image processing in multi-view stereo vision measurement for large forgings, including image quality evaluation, image contrast enhancement, and multi-view registration.As interfered by smoke and overlapped by forging press, the image quality of laser stripes should be evaluated to identify the clear image which can accurately denote forgings. Based on the analysis of the intensity profile for laser stripes, a mutual reference image quality evaluation method are put forward, which takes into account both the single image quality and the correlation of left and right images in a visual sensor. Experimental results of laser image evaluation show that this method can detect image quality accurately, as well as improve the efficiency of measurement.Because of the local brightness uneven of laser stripe, the forging feature cannot be extracted completely. An image contrast enhancement method combined the wavelet transform and contrast limited adaptive histogram equalization (CLAHE) is presented, laser stripes image enhancement and standard parts measurement experiments are conducted. Experimental results show that this method could improve the image quality, and lay the foundation for accurate extraction of forging features.Since the forging features are transformed into the same coordinate system, the relationship between two vision sensors must be calculated precisely. For this problem, a non-orthogonal laser grid pattern is designed, which projected onto the surface in overlapping region, and grid intersections of laser-aided pattern are regarded as registration feature points. Meanwhile, a high precision extraction algorithm based on region-of-interest (ROI) of registration feature points is proposed. Reconstructed results of these points are applied to calculate the transformation relationship using coarse and fine registration. The registration error is verified within 0.1%, in a field view of 8.6 m×5m, which satisfy the requirement of forgings measurement.Experiments in a forge are carried out, and the measured result of forging's average height is 1325.1 mm, which shows that the proposed multi-view stereo vision measurement method is valid for large forgings.