Study On 3D Shape Acquisition Methods For Objects With Specular Surface

As an important technology in the field of computer vision, 3D shape acquisition technology attracts much attention in the fields of factory automation and computer virtual reality in recent years. Many 3D measuring instruments have been developed such as arm-typed digitizers, laser range finders, stereo cameras, BGA inspector and so on. But these instruments are usually large scale and very expensive, and their dynamitic range and versatility are limited. There are many objects which have specular suface, such as the plastic and metal. The 3D shape measurement of high specular surface is not accurate and stable. The sepecular reflection causes serious error to the stereo matching due to its view-dependency. Meanewhile, the specular cause the camera CCD saturated and the information in this area missing. The proceeding time of the 3D shape acquisition method should be realtime for the practical application of factory automation. These bring challenge to the 3D shape acquisition technology.After a survey on 3D shape acquisition methods, we have done several creative works in the key problems of the 3D shape acquisition methods for the objects with specular surface. We study on not only the active, but also the passive methods. We propose new 3D shape acquisition methods which meet the real time requirement, have high accuracy, and are robust to noise in real application. The main contributions of this dissertation include:1. We present a novel high light spots removal algorithm which can remove the high light spots on the metal surface fast and efficiently. First, multi-light-sources method is adopted to reduce the large region of the highlight spots. We use several different light sources to get images and use the gradient to reconstruct an image. Multi-light-sources method cannot deal with the overlapped high light spots. For the residual specular spots, we use the total variation inpainting model to restore the diffuse region from neighborhood of the specular spots so as to eliminate them completely. The high light spots on the surface of the metal objects with complex texture can be removed effectively, and the physical property of the surface maintains well. The pyramid structure of our algorithm flow makes the innovation function efficiently.2. We propose a dense stereo algorithm based on census transform and 3-state dynamic programming (DP). Scanline-based DP algorithms are most efficient among global algorithms, but they are well-known to be affected by streak effect. To overcome this shortage, we improved the traditional 3-state DP algorithm by taking advantage of the combination of Canny edge detector and an extended version of sequential vertical consistency constraint. Experimental results show that our algorithm outperforms many state-of-the-art algorithms with similar efficiency on Middlebury College's stereo website. Besides, the algorithm is robust enough for image pairs with utterly different contrasts and noises.3. We present a simple, low cost and fast 3D shape measurement method. There is no limitation on the object's material and texture. We use a projector to project strip shifting pattern on the object, and a digital camera to record videos of the scene. The distortion of the strip shadow on the object is used to get the object's 3D information. A novel method is introduced to position the shadow edge accurately. This edge finding method overcomes the affect of inter-reflection and high light. Using this space-time information, we can calculate the pixels'3D coordinate. We improve black strip pattern to DeBruijn sequence to contatin more information in one image, and the processing time is reduced. The bad data are filtered by a post-processing 3D filter, which makes full use of the neighborhoods'geometric constrains and the view point constrain.