
2017年第36卷第9期
传感器与微系统(Transducer and Microsystem Technologies)
113
DOI:10.13873/J.10009787(2017)09011304
海事监管中无人机航拍图像快速拼接算法
欧阳帅,安博文,周凡,曹芳(上海海事大学信息工程学院,上海201306)
要:针对海事监管中航拍图像拼接生成大视场图像的时效性较低以及配准准确性不高的问题,提出摘
了--种快速高效的无人机(UAV)航拍图像拼接算法。根据海事监管辖区航拍图像特点缩小了角点搜索范围,通过设定自适应的梯度阔值和角点响应函数阀值筛选角点,通过局部最大角点响应函数值取舍准则实现了角点均匀化分布;采用基于相位相关的模板粗匹配方法和带有特征约束的RANSAC细匹配方法求出最优变换矩阵;利用人眼的视觉特性改进传统加权平均融合算法的加权因子使图像拼接过波自然。实
验结果表明:算法具有较好的自适应性,在拼接效率和准确率上较传统算法有了很大改善。关键词:航拍图像;角点检测:自适应阀值;图像配准;图像融合
中图分类号:TP317.4
文献标识码:A
文章编号:1000-9787(2017)09-0113-04
UAV aerial image fast mosaic algorithm in
maritimesupervision
OUYANG Shuai, AN Bo-wen, ZHOU Fan, CAO Fang
(College of Information Engineering,Shanghai Maritime University, Shanghai 201306, China)
Abstract: A fast and efficient image mosaic algorithm for unmanned aerial vehicle ( UAV) aerial image is proposed, which is able to cope with the problems of low speed and accuracy of generating large field of view image with aerial image mosaic in maritime supervision. First,scanning range of the angular point detection is narrowed according to the feature of aerial image. The gradient threshold and comer response function threshold are used to extract Harris cormer. Keeping the max value of local comer response function value is adopted to uniform the feature point distribution. After that,the improved phase correlation of template matching algorithm and improved RANSAC matching algorithm with constraint features are applied to obtain the optimal transformation matrix. Last, according to human visual characteristics, the improved factor of weighted average image fusion algorithm is applied to obtain a seamless image. Experimental results show that the algorithm has better adaptability. And the method overeomes the shortcoming of traditional mosaic methods,since the efficiency and accuracy of stitching are improved greatly
Key words: aerial image; angular point detection; adaptive threshold; image registration; image fusion
引
言
0
由于无人机(UAV){"航拍成像具有"站得高,看得远,跑得快”的优势,在船舶的动态管理、防污染监视、执法取证等方面具有其他方式不可比拟的优势,但无人机航拍的过程中受到飞行条件以及拍摄设备条件的限制,很难用-张图像将航拍目标区域的信息完全涵盖,因此,需要利用图像拼接技术获得完整的场景以解决实际需求。
目前,基于无人机图像拼接的方法主要有直接拼接法[3]和基于特征的方法。目前,航拍图像拼接算法大多基于特征点的方法"],主流的方法是尺度不变特征转换
(SIFT)特征点匹配法。虽然其具有高精度的特点,且对平移旋转具有较好的鲁棒性,但是非常耗时,无法满足实时要求5:。而Haris角点匹配法[6)]具有光强和旋转不变性,计算速度快,且角点具有显著的结构信息。郑兰等人提出了一种Harris自适应阔值角点提取方法,通过改进的随机抽样一致性(RANSAC)算法剔除误匹配点。杨宇博等人(* 提出了一种分块的Harris角点提取方法,配准采用分块逐步的RANSAC方法。
由于无人机沿河岸码头固定路线巡航航拍,图像中会出现较大一部分水域,该区域不存在角点特征,且水面容易
收稿日期:2016-09-23
·基金项目:国家白然科学基金资助项目(61171126);上海市重点支撑资助项目(12250501500):广西教育厅科研项目(YB2014207)