Scene image splicing method at presence of moving target

Abstract

The invention discloses a scene image splicing method at presence of a moving target. Firstly a scene image of the moving target is acquired by means of a video camera. Then a SIFT algorithm is utilized for extracting an image characteristic. Then characteristic matching between two images is finished by means of the extracted characteristic. Error characteristic matching is eliminated through a RANSAC algorithm, thereby obtaining correct characteristic matching. A geometric conversion model between images is established through a characteristic matching relationship, thereby finishing image registration. Finally fusion processing is performed on the registered image according to an image fusion method. According to the scene image splicing method, based on mobile direct linear transformation estimation and local constant linear fusion method, a splicing ghost caused by the moving target is eliminated through improving image registering precision, thereby improving quality and resolution of a spliced image. The scene image splicing method has advantages of simple and quick algorithm, low workload, high consistency, high efficiency and high registering precision. Furthermore the ghost caused by the moving target is eliminated. The scene image splicing method is particularly suitable for scene image splicing at the presence of the moving target.

Description

technical field [0001] The invention relates to a scene image mosaic method with moving objects, and belongs to the technical field of image mosaic in image panorama information. Background technique [0002] At present, panoramic image acquisition is an emerging research field and hot content of computer vision. There are two main ways to obtain panoramic images: (1) Directly use special wide-angle imaging equipment to capture images with a large enough horizontal angle at one time, such as non-linear optical imaging equipment such as fisheye optical lenses and convex reflective optical lenses, but the cost Higher, the resolution and viewing angle cannot be balanced, and the image will be severely distorted; (2) Through image stitching, a group of low-resolution or small viewing angle images with overlapping areas are stitched into a new image with high resolution and large viewing angle , its cost is low, and its effect is good. Therefore, image stitching methods are ver...

AI Technical Summary

Problems solved by technology

Some traditional image stitching methods have low registration accuracy and will produce parallax errors. Although some stitching "ghosts" caused by dynamic targets are also considered, they cannot solve the "ghosting" phenomenon well, so it will lead to stitching. image is not clear

[0005] For example, the currently widely used method with mature theory and simple calculation is based on the direct linear transformation estimation method and the linear fusion image stitching method. After estimating the global projection transformation matrix of the image with the direct linear transformation method, this one is used for all pixels in the image. The projection transformation matrix is ​​transformed without considering the correlation between pixels, so registration errors will occur

The simple linear fusion method takes the weight matrix corresponding to the transformed image (the element whose pixel value is not 0 corresponds to a weight of 1, otherwise it is 0), and in the three channels of the image, simultaneously transforms the transformed image and its corresponding weight matrix Multiply, add the multiplied results, and then divide by the sum of the weight matrix to get the stitched image after linear fusion. This method cannot eliminate the "ghost" caused by the moving target in the image, and the image clarity will be reduced. Greatly reduced

The above two problems make this image stitching method unable to obtain high matching accuracy and cannot effectively eliminate "ghosting", which is not suitable for scene image stitching with moving targets

Images