4K real-time video panoramic stitching method based on GPU acceleration

Abstract

The invention discloses a 4K real-time video panoramic stitching method based on GPU acceleration. The method comprises the following steps that hardware equipment takes streams and directly transmitsthe streams to a GPU video memory; calculating camera parameters and storing the parameters; calculating image brightness and chromaticity correction parameters; calculating a mask pattern, and performing image fusion on subsequent frames by using the mask pattern; mapping the images to a projection plane by using the calculated camera parameters, and superposing the overlapping areas of the twoimages together; performing compensation processing on the image by using the brightness and chroma correction parameters calculated in the step 3; and fusing the images by adopting an image pyramid and an optimal splicing technology to obtain a panoramic spliced image for display, output and storage. According to the method, the high-frequency detail information of the image is extracted by adopting a Gaussian multilayer pyramid method for fusion, and all operations are accelerated by using the GPU, so that the problems of ghosting and ghosting in the splicing process are solved, and the real-time requirement is met.

Description

Technical field [0001] The invention belongs to the technical field of image splicing, and particularly relates to a 4K real-time video panoramic splicing method based on GPU acceleration. Background technique [0002] Panoramic stitching technology is a technology that combines several overlapping images (which may be obtained at different times, different perspectives or different sensors) into a large seamless high-resolution image. Image stitching is collected and processed by images. , Image feature processing, image matching, model building, image fusion and other steps, and finally complete the mosaic of panoramic images. [0003] At present, there are many researches on video splicing technology at home and abroad, and the focus of the research is mainly on the two parts of single image registration and image fusion. There are few researches on the complex problems encountered in the process of continuous video stream splicing. The mainstream method of image registration i...

AI Technical Summary

Problems solved by technology

This method extracts less image detail information. When the brightness and chromaticity of the mosaic image are slightly different, the transition zone of the mosaic panorama is obvious and the visual effect is poor.

At the same time, the key to video panorama stitching technology is real-time performance. At present, most practitioners use CPU+GPU architecture to realize it. According to the performance of different graphics cards, they can only splice 2 to 4 channels of 2K video at most. , and there is no real-time compensation, resulting in uneven brightness and darkness of the mosaic image, and large color difference

In addition, considering that the image captured by the hardware image acquisition device may have vignetting (bright in the middle of the image and dark around it), there is no processing, which leads to the same problem in stitching panoramas

Images