Distortion Correction Of Large Field-Of-View Len And Image Stitching

Large field-of-view lens is have a wider field of view than ordinary lens, so it can contain more information in an image, but, the problems of serious distortion on the image is as the cost of wide field. In order to avoid affecting on the subsequent processing and the effect of image vision because of the distortion on image, distortion correction is needed. Now many image correction algorithms have been proposed, but they still with problem such like computing complex or generality. Video monitoring, traffic control, medical applications, intelligent city, and many other applications in computer vision field have an urgent need to use the large field of view images. Large field-of-view lens can solve some problems about the field in some degree, but there is also conflict of the resolution of image. It can use image mosaic technology to solve the contradiction well. In fact, a number of different mosaic algorithms produced due to the complexity situation of the applications, problems such as real time or accuracy of the algorithm appeared at the same time. This paper mainly studies the barrel distortion correction of large field of wide-angle lens and image mosaic based on feature points, to carry out research in the following aspects:(1) For the problem as the low accuracy of hemispherical model based algorithm, we proposes an improved hemispherical correction model, so that it can be more accurate and simple to achieve correction. Analysis application scenarios and the advantages and disadvantages of some classical correction algorithm, and decide the idea of correction algorithm as algorithm that based of parameter to implement correction of wide-angle lens distortion. We proposed an improved hemispherical correction model and algorithm. The algorithm achieves more ideal effect by introducing the concept of reverse mapping distortion rate, to avoid "empty" phenomenon; and addition a aparameter d which are change with r2on the focal length that is the distance between the spherical surface and the center, to reduce the imaging error caused by hemispherical model. The experimental results show that improved algorithm is better than the traditional hemispherical algorithm in precise and keep the speed of the traditional hemispherical algorithm. Because the algorithm uses only the focal length of the lens as a parameter, so it has good versatility.(2)According to the problems of classic Harris algorithm such as can’t detect corner of T, X and other types and rely on the experience to select parameters, an improved Harris corner extraction algorithm is proposed. The idea of the improved algorithm is:firstly, to lower the influence of edge response for points near the candidate interested points through taking the logarithm of the gradient magnitude, so that it can be more accurate positioning the corner; then according to the statistical character of the standard deviation to redefine the interested value function, to eliminate the experience value determined in the calculation of interested value and reduce the subjective influence on the algorithm, make it have a more accuracy positioning for corner detection. Experimental results show that, the algorithm can effectively detect the corner points of various types, and also with a higher repetition rate than Harris algorithm.(3)Based on the improved Harris corner extraction algorithm, we design and implement the point feature matching algorithm. This paper uses normalized cross-correlation (NCC) algorithm to realize coarse matching as the first step of matching, in order to remove unnecessary calculation of feature points matching. The second step of matching use random sampling algorithm consistency, namely the RANSAC algorithm, is the fine matching. This step eliminates the false matching in first step, and calculates the transformation matrix of image and its parameters. Lastly, using the weighted average fusion algorithm to complete the image mosaic and generated a wide field of view image of seamless.This paper improves the original image distortion correction and image mosaic algorithm, it provides new ideas and methods for correction distortion of wide-angle lens and image mosaic in computer machine vision., which can be reference for the related research and application, and also has scientific significance and research value.