桂林电子科技大学学报2024，Vol.44Issue(4)：340-347,8.DOI:10.16725/j.1673-808X.2023168
顾及动态目标的快速视频拼接
Fast video stitching with dynamic targets
摘要
Abstract
In view of the difficulty of real-time video splicing in scenes with dynamic targets,artifacts and dislocation are easy to oc-cur when moving objects pass through the joint of video splicing,and complex registration and fusion algorithms are difficult to meet the real-time requirements.In order to solve this problem,this paper proposes a fast video splicing algorithm that takes dynam-ic targets into account.For the collected video frames,ORB algorithm was used to match sparse feature points between frames,YOLOv5 was used to segment possible moving targets(cars and pedestrians)in the left and right frames,and then combined with optical flow method to determine whether the entire target area is a moving target area and remove the influence of moving targets,and a single mapping H-array estimation was performed between frames based on the pair of feature matching points in the static area.The highly accurate H-array is obtained.Finally,when image fusion is carried out through the optimal suture line,the suture line can avoid the moving target area based on the previously detected moving target area,and the dynamic target area is updated on the left and right sides respectively.Finally,the new algorithm is tested in detail,and the average stitching speed can reach 61 ms ev-ery two frames(1 280×720 resolution),which has a faster stitching speed than APAP and ELA algorithms,and finally achieves a fast video stitching that can take into account the influence of dynamic objects.关键词
YOLOv5/稀疏特征点匹配/视频拼接/最佳缝合线/运动目标检测Key words
YOLOv5/sparse feature point matching/video stitching/best seam-line/moving target detection分类
信息技术与安全科学引用本文复制引用
张文涛,郑曦尧,彭智勇..顾及动态目标的快速视频拼接[J].桂林电子科技大学学报,2024,44(4):340-347,8.基金项目
广西自然科学基金(2020GXNSFAA19091) （2020GXNSFAA19091）
桂林电子科技大学研究生教育创新计划(2023YCXS222) （2023YCXS222）
