Multi-scale cylindrical projection panorama image generating method for video monitoring

Abstract

The invention relates to the technical fields of image processing and video processing and provides a multi-scale cylindrical projection panorama image generating method for video monitoring. The method comprises the following steps: intercepting a preset overlapping region of a to-be-spliced image, thereby acquiring a cylindrical projection image; establishing sub-images of the image under various scales, searching feature points by adopting an improved Harris method, and solving feature vectors under the scales; searching the matched feature point pairs, filtering, solving the optimized homography matrix, coordinates of the splicing joints and the actual overlapping region of to-be-spliced images; and adjusting the chroma so as to eliminate the splicing joints, thereby obtaining the final panorama image. The method disclosed by the invention has the great advantages that the image edge distortion caused by the shooting angle of a camera is eliminated, the processing speed and the algorithm stability are improved by adopting a mode of presetting the overlapping region, the solved matching points are filtered, the image color difference is adjusted by adopting a gradual coefficient adjustment mode, and the final joint-free panorama image with consistent chroma is formed.

Description

technical field [0001] The present invention relates to the technical field of image processing and video processing, in particular to a multi-scale cylindrical projection panoramic image generation method for video monitoring. Background technique [0002] The key technology of image stitching is to accurately find the overlapping area and range between image sequences, and realize the correspondence model between corresponding points, that is, image registration. According to the currently proposed image registration algorithms, image stitching techniques can be divided into three categories: methods based on transform domain, methods based on image features and methods based on image grayscale. The method based on the transformation domain is difficult to deal with the rotation and scaling of the lens, and requires a large overlapping ratio between the images to be registered, and searches the entire image space at the same time, which is computationally expensive. The m...

AI Technical Summary

Problems solved by technology

For example: "Creating full view panoramic image mosaics and environment maps" published by Szeliski R on Proceedings of the 24th annual conference on Computer graphics and interactive techniques to manually determine at least 4 pairs of feature points, and construct optimization based on the brightness error of feature point pixels Function to estimate the point transformation relationship between images, but the number of feature points obtained by this method is too small, and the registration effect is not ideal

[0004] After searching, it was found that in Jiang Daihong et al.’s patent "A Quick Mosaic Method for Mine Monitoring Images", the Harris algorithm was used to extract the corner points and obtain the SIFT descriptors of the corner points. However, when obtaining the SIFT descriptors in this method, the description The dimension of the character is large, the required calculation time and storage space are large, and it cannot be overcome when the edge of the image is distorted due to the angle of the camera.

[0005] The above methods do not take into account the correction of the edge distortion of the image due to the shooting angle, and the accuracy of image registration and fusion as well as the calculation speed need to be considered comprehensively

Images