摘要
针对大型机场补盲监视、中小机场远程塔台监视的全景监视需求,提出了一种基于多路视频与GPU CUDA并行加速的全景监视计算框架。首先实时采集多路高清网络摄像机视频,针对视频间光照不一致的问题,提出了一种基于HSV色彩空间与全局优化函数的动态光照一致化算法。为提升全景生成质量,设计了一套基于RANSAC与FAST特征的变换模型自动估计算法,并通过动态规划算法实现基于拼接缝的全景图缝合。对大规模、高并发的视频图像数据进行CUDA并行加速处理。提出的方法通过国内某机场应用示范,图像拼接具有精度高、分辨率高、实时性好等特点,能满足机场补盲、远程塔台等全景监视的需求。
Aiming at the panoramic surveillance requirements of large-scale airport blind surveillance and remote tower surveillance of small and medium-sized airports,this paper has proposed a panoramic surveillance computing framework based on multi-channel video and GPU CUDA parallel acceleration.First,multi-channel high-definition network camera videos were captured in real time.In order to solve the problem of inconsistent lighting between vide-os,a dynamic lighting consistency algorithm based on HSV color space and global optimization function was proposed.Then,a set of automatic estimation algorithm of transformation model based on RANSAC and FAST features was designed to improve the quality of panorama generation,and the panorama stitching based on mosaic seam was realized through dynamic programming algorithm.Finally,CUDA parallel acceleration computing architec-ture was performed on large-scale,high-concurrency video image data.The method proposed in this paper has been applied and demonstrated in a domestic airport.It has the characteristics of high image stitching accuracy,high pano-ramic resolution,and good real-time performance.It can meet the needs of panoramic surveillance,such as airport blind compensation and remote towers.
作者
吴岳洲
傅强
郭康
张先浩
WU Yue-zhou;FU Qiang;GUO Kang;ZHANG Xian-hao(School of Computer Science,Civil Aviation Flight University of China,Deyang Sichuan 618307,China)
出处
《计算机仿真》
北大核心
2023年第5期130-134,149,共6页
Computer Simulation
基金
国家重点研发计划资助(2021YFF0603904)
中国民用航空飞行学院科研基金重点项目(ZJ2022-004)。
关键词
图像特征
图像配准
并行加速
图像融合
全景监视
Image features
Image registration
Parallel acceleration
Image fusion
Panoramic surveillance
