Fish eye monitoring image splicing method based on combination and matching of multiple feature points

Abstract

The invention discloses a fish eye monitoring image splicing method based on combination and matching of multiple feature points. The method comprises the following steps: carrying out distortion correction on an image by applying a barrel-shaped distortion model and Hough line detection; extracting N feature points from the selected areas of the two images and respectively solving feature descriptors of the feature points; pre-matching the N feature points by using a violent matching method or a fast nearest neighbor search method; performing confidence ranking on the pre-matching pairs of the N feature points; selecting two pairs of matching pairs with the highest confidence coefficient of each feature point to form an optimization set, searching for and reserving four groups of optimalmatching pairs, and eliminating other matching pairs; selecting a matching pair with a sub-high confidence coefficient in the obsoleted feature points, adding the matching pair into the optimization set, and performing optimization again until a termination condition is met; and calculating a homography matrix according to the four groups of matching pairs obtained through optimization, transforming the image and realizing splicing. According to the method, the homography matrix is calculated through the combination of the matching pairs among the multiple feature points, so that the selectionaccuracy of the matching pairs is remarkably improved, and then effective splicing of two monitoring images is realized.

Description

technical field [0001] The present invention relates to the technical field of image stitching, in particular to a fisheye surveillance image stitching method based on combination and matching of various feature points, in particular to non-parametric correction, image feature point matching and image stitching methods of fisheye surveillance images. Background technique [0002] Fisheye monitoring images have serious distortion compared with ordinary images. Generally, it is necessary to calibrate the camera to obtain the distortion parameters of the camera, and then perform distortion correction on the images captured by the camera. But for the situation where the camera parameters cannot be collected, the distortion correction of the image becomes difficult. [0003] The general implementation process of image stitching technology is: extract some feature points of the image, match the feature points, calculate the homography matrix through feature point matching, deform ...

AI Technical Summary

Problems solved by technology

[0004] However, the general image stitching process cannot achieve a good stitching process for surveillance image stitching under two different viewing angles that are far apart, especially when there are similar objects in the overlapping area (such as zebra crossings, repeated patterns on the sidewalk, dense trees, etc.) Wait)

Because the matching of feature points will have great interference, the correct matching of most feature points cannot be achieved, which leads to the inability to obtain the correct homography matrix for the image, thus making it impossible to achieve splicing

Images