Research On The Application Of Computer Vision On Coal Stocktaking Of Thermal Power Plant

With the development of computer technology, the domain of computer vision has drawn more and more attention from people. Stereo vision mainly studies the application of two-dimensional (2D) projection image to restore the three-dimensional (3D) scenery world, which is one of the main tasks of computer vision technology. Computer 3D reconstruction theory and its technical application play an important role in the industrial, scientific and other fields.According to the specific requirement of the thermal power plant for measuring the volume of coal stockpile, from the angle of the theory and the realization of technology, this paper studies the problems of stereo vision technology, which include the image pretreatment, the image feature extraction, the matching based on image feature and image segmentation, and the 3D reconstruction.The extraction of the image essential features is the foundation of the computer 3D reconstruction. According to the analysis of the image linear features, this paper puts forward an extraction algorithm which combines point, line and region together. After the extraction of features of the edge as well as its corresponding point and line, this algorithm analyzes the relevance of these extracted features, and increases the amount of feature point and line by applying polar, providing more accurate features for the future image matching. This method not only makes use of the relevance of the futures, but also the mathematic method such as polar, therefore these increased features are of high reliability.When the introduction of the segmentation is made, the image has been divided into a smoother region and a region in which the illumination change is fiercer before matching the image. To the region which has been divided, the threshold is calculated repeatedly for the purpose of obtaining ideal segmented region. And when the region segmentation is made once more, the influence of the extracted feature line on segmentation is taken into account at the same time.The image matching is the core of the 3D reconstruction problem. This paper proposes a kind of matching of the integration of binocular vision and monocular vision, which is based on the image features and the image segmentation. This algorithm allows for not only the features of relevant point and line, but also the features of region at the time of image feature extraction and relevant region matching. When matching, in accordance with the known point, line and the segmented small region, the matching proceeds effectively by synthesizing their features and improves success rate and accuracy of the matching. In respect of 3D depth restoration, according to the depth value estimated by binocular vision which is calculated in the light of known feature matching, and then according to the formula of the probability theory, combining the algorithm which estimates the depth value by monocular vision on the basis of the partial feature and overall feature of the image, the fusion of the algorithm of depth calculation is put into practice in order to obtain more accurate depth value to carry out the 3D reconstruction.