基于跑道边界跟踪与图像融合的视景增强研究
发布时间：2018-10-25 14:36
【摘要】：针对雨、雪、雾等恶劣天气导致的低能见度对飞机着陆安全性的不利影响,开展了飞行员着陆视觉增强相关研究工作。通过跑道检测、跟踪与多源传感器信息融合技术进行面向飞行员着陆的视觉增强研究,以期显著提高低能见度下机场跑道环境的可视性。本课题提出“基于跑道边界跟踪与图像融合的视景增强算法”,该方法能够有效利用机载前视红外与可见光视频图像信息,显著提高低能见度条件下飞行员的视觉感知能力。算法的主要优点是:能够更直观的向飞行员指示跑道边界等信息;通过图像融合能够综合异源传感器的互补信息,使视景增强效果更接近飞行员日常视景;本课题提出的视景增强算法具有较高的实时性。本文主要工作由以下部分组成:首先,分析了视景增强(EVS,Enhanced Vision System)技术的出现背景,并对视景增强技术的的理论研究与应用研究现状就行了介绍,对现在主要研究成果进行了梳理。指明了现有视景增强技术的不足,并介绍了本文所提算法。其次,研究了机载前视红外(FLIR)视频首帧图像的跑道检测问题。针对现有基于单特征的跑道检测算法误检率高的问题,提出了基于多特征融合的跑道检测算法。该算法融合了直线、灭点、先验知识等信息联合进行跑道检测,极大地降低了误检率,提高了检测精度。其中研究了LSD(Line Segment Detector)直线检测、基于Gabor特征的灭点检测、基于直线特征的灭点检测等算法。然后,研究了机载前视红外(FLIR)视频序列中机场跑道的跟踪问题。针对跑道区域的大跨度特性和明显的边界约束性,提出了基于多采样点联合定位的跑道跟踪算法。该算法充分利用跑道区域的稀疏特性,即跑道区域由跑道的边界直线与图像边界共同决定,而直线的跟踪又可以由直线上任意两点的跟踪实现。故而,通过对四个采样点的跟踪便实现了对整个跑道区域的精确定位。实验表明该算法具有较高的实时性与跟踪精度。再者,研究了基于图形学优化理论的区域分割算法与基于多尺度分析的图像融合算法。针对图像多尺度分析只能针对矩形区域的特点,对任意多边形跑道区域进行矩形化分割。采用迭代的方式对多边形区域进行分割直到达到最小阈值。其后,采用小波变换(WT,Wavelet Transform)对红外与可见光视频图像的跑道区域(ROI,region of interest)进行融合,而对跑道以外的区域采用简单的加权策略进行融合。这样就在保证跑道区域融合性能的同时,满足了视景增强实时性的要求。最后,在完成视景增强系统设计的基础上,研究了基于边缘特征的障碍物尺寸估计问题。在飞机着陆过程中跑道中的微小障碍物都将对飞机安全构成极大威胁,障碍物往往在可见光图像中不易察觉,而在红外成像中会具有明显的轮廓特征。算法首先对候检区域进行显著性(Salience)检测,其后对显著目标进行基于边缘特征的尺寸估计。
[Abstract]:Aiming at the adverse effects of poor visibility caused by severe weather such as rain, snow and fog on the landing safety of aircraft, the related research work on visual enhancement of pilot landing was carried out. Based on runway detection tracking and multi-source sensor information fusion the visual enhancement for pilot landing is studied in order to improve the visibility of runway environment in low visibility. In this paper, a visual enhancement algorithm based on runway boundary tracking and image fusion is proposed. This method can effectively utilize airborne infrared and visible image information, and improve the visual perception ability of pilots under low visibility. The main advantages of the algorithm are that it can direct the runway boundary to the pilot more intuitively, and the complementary information of the heterogeneous sensor can be synthesized by image fusion, so that the visual enhancement effect is closer to the daily scene of the pilot. The scene enhancement algorithm proposed in this paper has high real-time performance. The main work of this paper consists of the following parts: firstly, the background of scene enhancement (EVS,Enhanced Vision System) technology is analyzed, and the present situation of theoretical research and application research of visual enhancement technology is introduced, and the main research results are summarized. The shortcomings of the existing scene enhancement techniques are pointed out, and the algorithms proposed in this paper are introduced. Secondly, the runway detection problem of the first frame image of airborne forward looking infrared (FLIR) video is studied. Aiming at the problem of high false detection rate of the existing runway detection algorithm based on single feature, a runway detection algorithm based on multi-feature fusion is proposed. The algorithm combines the information of straight line, vanishing point and prior knowledge for runway detection, which greatly reduces the false detection rate and improves the detection accuracy. The algorithms of LSD (Line Segment Detector) line detection, vanishing point detection based on Gabor feature and vanishing point detection based on line feature are studied. Then, the tracking problem of airport runway in airborne forward-looking infrared (FLIR) video sequence is studied. In view of the large span characteristic and obvious boundary constraint of runway region, a runway tracking algorithm based on joint location of multiple sampling points is proposed. The algorithm makes full use of the sparse property of the runway region, that is, the runway region is determined by the boundary line of the runway and the image boundary, and the tracking of the straight line can be realized by the tracking of any two points on the line. Therefore, by tracking the four sampling points, the accurate location of the entire runway area is realized. Experiments show that the algorithm has high real-time and tracking accuracy. Thirdly, the region segmentation algorithm based on graphic optimization theory and the image fusion algorithm based on multi-scale analysis are studied. In view of the multi-scale analysis of the image, only the characteristics of the rectangular region can be used to segment the arbitrary polygonal runway region. The polygon region is segmented by iterative method until the minimum threshold is reached. Subsequently, wavelet transform (WT,Wavelet Transform) is used to fuse the runway region (ROI,region of interest) of infrared and visible video images, while the region outside the runway is fused by a simple weighted strategy. In this way, the performance of runway area fusion is guaranteed, and the requirement of real-time scene enhancement is satisfied. Finally, based on the design of scene enhancement system, the problem of obstacle size estimation based on edge features is studied. The small obstacles in the runway will pose a great threat to the safety of the aircraft during the landing process. The obstacles are often difficult to detect in the visible image and have obvious contour characteristics in the infrared imaging. The algorithm firstly detects the significant (Salience) of the waiting area, and then estimates the size of the salient target based on the edge feature.
【学位授予单位】：西北工业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：V241;TP391.41
本文编号：2293984
[Abstract]:Aiming at the adverse effects of poor visibility caused by severe weather such as rain, snow and fog on the landing safety of aircraft, the related research work on visual enhancement of pilot landing was carried out. Based on runway detection tracking and multi-source sensor information fusion the visual enhancement for pilot landing is studied in order to improve the visibility of runway environment in low visibility. In this paper, a visual enhancement algorithm based on runway boundary tracking and image fusion is proposed. This method can effectively utilize airborne infrared and visible image information, and improve the visual perception ability of pilots under low visibility. The main advantages of the algorithm are that it can direct the runway boundary to the pilot more intuitively, and the complementary information of the heterogeneous sensor can be synthesized by image fusion, so that the visual enhancement effect is closer to the daily scene of the pilot. The scene enhancement algorithm proposed in this paper has high real-time performance. The main work of this paper consists of the following parts: firstly, the background of scene enhancement (EVS,Enhanced Vision System) technology is analyzed, and the present situation of theoretical research and application research of visual enhancement technology is introduced, and the main research results are summarized. The shortcomings of the existing scene enhancement techniques are pointed out, and the algorithms proposed in this paper are introduced. Secondly, the runway detection problem of the first frame image of airborne forward looking infrared (FLIR) video is studied. Aiming at the problem of high false detection rate of the existing runway detection algorithm based on single feature, a runway detection algorithm based on multi-feature fusion is proposed. The algorithm combines the information of straight line, vanishing point and prior knowledge for runway detection, which greatly reduces the false detection rate and improves the detection accuracy. The algorithms of LSD (Line Segment Detector) line detection, vanishing point detection based on Gabor feature and vanishing point detection based on line feature are studied. Then, the tracking problem of airport runway in airborne forward-looking infrared (FLIR) video sequence is studied. In view of the large span characteristic and obvious boundary constraint of runway region, a runway tracking algorithm based on joint location of multiple sampling points is proposed. The algorithm makes full use of the sparse property of the runway region, that is, the runway region is determined by the boundary line of the runway and the image boundary, and the tracking of the straight line can be realized by the tracking of any two points on the line. Therefore, by tracking the four sampling points, the accurate location of the entire runway area is realized. Experiments show that the algorithm has high real-time and tracking accuracy. Thirdly, the region segmentation algorithm based on graphic optimization theory and the image fusion algorithm based on multi-scale analysis are studied. In view of the multi-scale analysis of the image, only the characteristics of the rectangular region can be used to segment the arbitrary polygonal runway region. The polygon region is segmented by iterative method until the minimum threshold is reached. Subsequently, wavelet transform (WT,Wavelet Transform) is used to fuse the runway region (ROI,region of interest) of infrared and visible video images, while the region outside the runway is fused by a simple weighted strategy. In this way, the performance of runway area fusion is guaranteed, and the requirement of real-time scene enhancement is satisfied. Finally, based on the design of scene enhancement system, the problem of obstacle size estimation based on edge features is studied. The small obstacles in the runway will pose a great threat to the safety of the aircraft during the landing process. The obstacles are often difficult to detect in the visible image and have obvious contour characteristics in the infrared imaging. The algorithm firstly detects the significant (Salience) of the waiting area, and then estimates the size of the salient target based on the edge feature.
【学位授予单位】：西北工业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：V241;TP391.41
【参考文献】
相关期刊论文 前3条
1 席亮;;基于量子力学和拉普拉斯金字塔的图像融合方法[J];计算机与现代化;2015年04期
2 冯伍;张俊兰;苗秋瑾;;几种典型边缘检测算子的评估[J];电子设计工程;2011年04期
3 杨顺辽;卢凌;杜宗岗;梁军;李晟;;SAR图像中机场跑道的自动识别研究[J];武汉理工大学学报(交通科学与工程版);2006年01期
，本文编号：2293984
本文链接：https://www.wllwen.com/kejilunwen/hangkongsky/2293984.html
