Research On Acquisition Processing And 3D Reconstruction For Structured Light Range Image

At present, structured light measurement and range image processing have been widely used in many fields. In this dissertation, these two technologies are integrated seamlessly to measure and restore 3D objects. Studies have been made in many aspects such as: structured light measurement, range image interpolation; range image filtering, regional segmentation of range image and 3D reconstruction of range image.Fundamental theories of traditional structured light measurement were analyzed and a multi-line structured light coding method with marked strip was used. This method combines the characteristics of both gray and color encoding. And it can decrease error rate in gray decoding and prevent mistakes caused by diffraction and interfere in color coding. Multi-line structured light measurement system was built based on the proposed coding method. Measuring results showed that this system can meet with design specifications.When multi-line structured light are used in 3D measurement, data loss occurred to some extent. Aiming at charasteristics of structured light range image, two interpolating methods were presented by analyzing traditional image interpolation. One is called self-adaptive weighted interpolation used in processing range image in vertical strip region, while the other is average value interpolation applied to process range image in non-vertical strip region. Experimental results showed that range data are continuous and dense after interpolating range images with these two methods.Classic edge detecting techniques can not find the boundaries of corrugated range images due to the slow variation of range values. To solve this problem, two range image segmentation methods were proposed. One is based on edge information fusion of normal components in which two normal vectors are extracted and fused. By combing edge detection and morphology, special detection operators were introduced to segment images with jumping and corrugated boundaries. Segmented regions are nearly ideal.It will be time consuming and low accuracy to triangulate and reconstruct range images directly because of its large data size and complex topology structure. Using isomorphic neural network, a partition reconstruction method was presented. Firstly, range image was divided into some regions with the help of image segmentation. Secondly, isomorphic RBF neural network was applied to every region by training samples with different resolutions, aimed at adapting to different structure characteristics in regional reconstruction. In the end, all regions are combined and reconstructed.Finally, 3D measurement and restoration of regular geometrical bodies and head sculptures were carried out separately. Experimental results show methods proposed in this dissertation are effective and robust to 3D object reconstruction.A Framework from multi-line structured light measurement to range image processing and 3D object reconstruction was proposed, which has theoretical and practical meanings to 3D objection measurement and restoration.