陆海一体数字孪生智慧渔场平台中的视频图像拼接方法

Video Image Stitching Method for the Land-Sea Integrated Digital Twin Smart Fish Farm Platform

关键词:  图像处理，图像拼接，图像配准，图像融合，智慧渔场，数模融合

Key words:  image  processing, image  stitching, image  alignment, image  fusion, smart  fishery, data-model  integration

摘要:

在智慧渔场的管理中，当面对较大空间范围的全局实时监测需求时，监控视频图像拼接可以解决由于监控视频零散造成应对不及时甚至误判的问题。针对智慧渔场视频图像拼接过程中图像配准阶段耗时长、特征匹配率低、图像特征纹理弱等问题，设计出一种BRISK+GMS组合式图像配准算法。在BRISK快速提取特征点的基础上，使用结合双向匹配策略的GMS-RANSAC算法筛选优秀特征点对，使得后续的图像融合阶段更加平滑自然。为了解决图像融合阶段出现的拼接缝、重影等问题，设计出一种结合改进渐入渐出算法的最佳缝合线算法，优化最佳缝合线算法中的能量函数，引入HSV色彩空间中的色彩饱和度代替色差强度，使得在最小色差区域进行拼接，同时优化渐入渐出融合算法中的权值函数，使图像融合过程中图像过渡更加平滑自然，消除缝合线以及重影现象。通过实验验证了图像配准阶段和图像融合阶段算法的在视频图像拼接系统中的可行性和实用性。实验结果表明，相对于传统的拼接融合算法，本文算法一定程度上在保证消除拼接缝和重影等问题的同时，保证了视频图像拼接的实时性。

Abstract:

In the management of smart fish farms, where there is a need for comprehensive real-time monitoring over a large spatial area, video image stitching addresses issues of delayed response or even misjudgment resulting from scattered surveillance videos. Addressing issues such as prolonged registration time and low feature matching rates during the video image stitching process, a BRISK+GMS combined image registration algorithm is proposed. Building upon the fast feature point extraction of BRISK, the GMS-RANSAC algorithm with a bidirectional matching strategy is used to filter out high-quality feature point pairs, thereby ensuring a smoother and more natural subsequent image fusion stage. To address issues such as seam artifacts and ghosting in the image fusion phase, a novel Best Seam Line algorithm, incorporating an improved fade-in/fade-out approach, is proposed. The algorithm optimizes the energy function in the Best Seam Line algorithm, introducing saturation from the HSV color space as a replacement for color intensity. This ensures seamless stitching in regions with minimal color differences. Additionally, the weighting function in the fade-in/fade-out fusion algorithm is optimized, facilitating a smoother and more natural transition during the image fusion process, eliminating seam artifacts and ghosting phenomena. The feasibility and practicality of the algorithms in both the image registration and image fusion stages were validated through experiments in a video image stitching system. Experimental results indicate that, relative to traditional stitching and fusion algorithms, the proposed approach to some extent ensures real-time performance in video image stitching while effectively addressing issues like seam artifacts and ghosting.