基于特征的无人机载视频运动目标快速检测方法  

A fast detection method for moving targets in UAV video based on the feature extraction

作　　者：谭熊[1] 王晶磊 孙一帆 TAN Xiong;WANG Jinglei;SUN Yifan(Strategic Support Force Information Engineering University, Zhengzhou 450001, China)

机构地区：[1]战略支援部队信息工程大学,郑州450001

出　　处：《国土资源遥感》2021年第2期27-32,共6页Remote Sensing for Land & Resources

基　　金：高分重大专项项目(编号:GFZX04040202-04)资助。

摘　　要：针对无人机载视频序列影像动态背景以及常用检测方法处理效率不高等问题,提出了一种基于特征的无人机载视频运动目标快速检测方法。该方法主要包括预处理、ORB特征提取(oriented fast and rotated BRIEF,ORB)、渐进一致采样(progressive sample consensus,PROSAC)特征精匹配、全局运动估计与全局运动补偿、运动目标初检测和形态学后处理等6个步骤。通过2组无人机载视频数据实验研究表明:对不同传感器获取的不同帧率和不同尺寸大小的视频序列影像,采用本方法的运动目标检测结果较好,计算效率最高,且能够满足实时处理需求。Aiming at tackling the problem of the low efficiency of common detection methods and the dynamic background of UAV-based video sequences imagery,this paper proposes a fast detection method for moving targets in UAV video based on the feature extraction.The method mainly includes six steps,i.e.,preprocessing,ORB feature extraction,PROSAC feature fine matching,global motion estimation and global motion compensation,initial detection of moving target and morphological post-processing.The results of two video experiments carried by UAVs show that the moving target detection results of the proposed method in this paper are better,the computational efficiency is the highest,and hence this method can meet the requirements of real-time processing.

关 键 词：视频序列影像 ORB特征 PROSAC精匹配 运动目标检测 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]