顾及动态目标的快速视频拼接  

Fast video stitching with dynamic targets

作　　者：张文涛[1] 郑曦尧 彭智勇[1] ZHANG Wentao;ZHENG Xiyao;PENG Zhiyong(School of Optoelectronic Engineering,Guilin University of Electronic Technology,Guilin 541004,China)

机构地区：[1]桂林电子科技大学光电工程学院,广西桂林541004

出　　处：《桂林电子科技大学学报》2024年第4期340-347,共8页Journal of Guilin University of Electronic Technology

基　　金：广西自然科学基金(2020GXNSFAA19091);桂林电子科技大学研究生教育创新计划(2023YCXS222)。

摘　　要：动态目标的场景实时视频拼接是视频拼接中的难点,当运动对象经过视频拼接缝时易发生伪影和错位,而复杂的配准和融合算法又难以满足实时性要求。为了解决该问题,提出了一种顾及动态目标的快速视频拼接算法。针对采集的视频帧,通过ORB算法进行帧间稀疏特征点匹配,YOLOv5分割出左右两帧中的可能运动目标(汽车和行人),再结合光流法判断整个目标区域是否为运动目标区域,然后去除运动目标影响,基于静止区域的特征匹配点对进行帧间单映射H阵估计,得到高精确H矩阵。最后通过最佳缝合线进行影像融合时,基于前面检测出的运动目标区域,使缝合线规避运动目标区域,同时在左右两边分别针对动态目标区域进行更新。算法测试结果表明,每两帧(分辨率1280×720)平均拼接速度可达61ms,拼接效果好于APAP和ELA算法,且具有更快的拼接速度,最终实现了能顾及动态目标影响的快速视频拼接。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(1280×720 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 

分 类 号：TN911.74[电子电信—通信与信息系统] TP391.41[电子电信—信息与通信工程]