The Application Technology Research Of Panoramic Image Stitching

With the continuous development of modernization construction, China's industrial waste water emissions continue to increase and the waste water treatment is facing great pressure. The cleaning and recycling of sundries in the pool of industrial waste water is an important link in the process of waste water treatment. The underwater intelligent robot as a carrier of the operation in complex underwater environment, has good visual perception ability and is highly intellectualized, are widely used to the operation in complex and high risk underwater environment. Therefore, the research in visual processing technology of underwater intelligent robot has great practical value and is of important theoretical significance. In this paper, the visual image processing system of the underwater intelligent robot based on monocular vision is studied, mainly including the following aspects of content:Aiming at the influence of the existence of various noise in underwater environment as well as the light scattering and refraction effect on the quality of underwater image, this paper proposes a fuzzy enhancement method based on minimum relative entropy criterion to enhance the image. First of all, mapping the image to fuzzy domain according to membership function , then according to the minimum relative entropy criterion to determine the transit point of fuzzy transformation to enhance the image in fuzzy domain. Through operating results contrast with the traditional image enhancement methods, this method can effectively improve the quality and the contrast of underwater image.The next is to segment the target image. The uneven underwater illumination makes the distribution of image gray level uneven, seriously affected the image segmentation effect.Aiming at this problem, combining with the image color space, this paper puts forward a image segmentation method based on tone and the maximal difference between categories.Inserting from H of the underwater visual image's HSI color space, according to the H value's differences between its target and background pixel, to make transformation on the gray level to increase the gray level differences, and then segmenting the image according to its threshold. And then verified by experimental simulation in this paper, the proposed algorithm can effectively restrain the influence of uneven underwater illumination on image segmentation, and the traditional threshold segmentation results are compared with and analyzed.At the link in target classification and recognition, this paper reconstructs 6 new invariant moments with the feature of rotation, translation and scaling invariance on the basis of traditional Hu invariant moments , to extract the shape of the target image as the basis of classification. In view of the traditional BP neural network's slow convergence speed, easy to fall into local minimum and other shortcomings, in this paper, we combination the particle swarm optimization algorithm and BP neural network, using particle swarm algorithm in neural network's training to design the BP neural network classifier based on particle swarm algorithm. And this improves the training speed of neural network, and through the simulation experiments shows the advantages in the target recognition rate that the classifier has over traditional BP neural network classifier.In the end, we determine the three-dimensional position of underwater target object.Traditional monocular or binocular vision positioning system requires complex camera calibration, combined with BP neural network in this chapter, on the basis of the principle of binocular vision positioning we design a new kind of monocular 3-D positioning method which is based on the BP neural network, successfully avoids the camera calibration and reduces the system complexity.