Real-time correction method and system for self-learning multipath image fusion

Abstract

The invention provides a real-time correction method and a real-time correction system for self-learning multipath image fusion. The Real-time correction method comprises the steps of: extracting and matching contours and object features in a field of view at first, learning to generate a matching template suitable for rapid fusion, and establishing a fusion matrix of initial evaluation blocks according to pre-segmentation macro blocks of an image; secondly, acquiring multipath image data at multiple moments in one step and fusing the multipath image data with the matching template according to fixed frame rate pace; evaluating the fused image, and judging whether the fusion successes; and minimizing the minimal macro block in an original image feature region by one level if the image fusion successes, and generating a template for further improving precision and differentiation degree for standby. By means of feature matrix universality of the plurality of adjacent frames, the real-time correction method and the real-time correction system adopt different known feature matrixes and correction methods according to the judgment of tendency of feature matrix variation, fuse and splice the images in a multipath real-time video stream quickly, realize the real-time processing of video images, and enable the multipath real-time image video stream to be spliced stably and continuously.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a real-time correction method and system for self-learning multi-channel image fusion. Background technique [0002] At present, the wide-angle lens images on the market will have different degrees of image distortion with the increase of the angle, especially the distortion of the spherical lens becomes more serious. The multi-eye array camera is used to combine multiple images to obtain a low-distortion ultra-wide view. The method of the present invention belongs to the calculation method of real-time stitching and correction of multi-array images. [0003] When stitching multiple images, there are currently two main methods: [0004] 1) By extracting and comparing the boundary feature information in the two images near the splicing, the pixels based on these similar feature information are selected as the median of the fitting curve to fuse multiple images during fit...

AI Technical Summary

Problems solved by technology

[0017] The methods in the above literature belong to the target image quality improvement of image post-processing, and have nothing to do with splicing and fusion itself.

Secondly, using two teaching aids to shoot two sets of images cannot be applied to multi-channel real-time video stream image acquisition

Images