【作者】 郭艳颖；

【导师】 杨国庆；

【作者基本信息】 南京航空航天大学 ， 计算机应用技术， 2012， 博士

【摘要】 飞机泊位自动引导的实施对提高机场信息化和自动化水平至关重要，基于视觉的泊位自动引导方法因为具有信息丰富、效果直观及成本低等优点一直受到国内外学者的关注。利用图像处理技术检测飞机边缘轮廓，识别出飞机机型进而跟踪，完成自动引导关键技术算法研究。本文主要针对特殊天气下利用图像处理手段，重点研究了满足特殊天气下泊位图像去噪等预处理算法、泊位飞机的轮廓检测、识别和跟踪等自动引导的关键技术。首先，深入研究了特殊天气环境下图像增强和去噪等预处理算法。针对光照过强、夜间等特殊环境光照条件导致对比度不均衡问题，提出了分段变换方法，此算法实现简单，运算速度快，提高了图像对比度，能够满足系统的实时性要求；针对雾霾天气时能见度降低，造成检测识别困难，提出了基于暗原色优先的形态学去雾算法，有效提高目标的清晰度，保留图像细节边缘，去雾效果自然逼真，同时满足实时性和鲁棒性要求。其次，针对目标飞机分割问题研究了边缘检测算法和阴影分割算法。在边缘检测方面，针对边缘的细节及噪声问题，提出了基于自适应权重边缘检测算法；在图像分割方面，运动物体的阴影被标记为前景会增加目标识别和跟踪算法的复杂度，提出了一种将形态学的幂变换与无边界主动轮廓线模型结合的阴影分割算法，牺牲很短的数学形态学平滑的时间，换来了主动轮廓的快速收敛，从而减少了总的运算时间，并且抑制了泊位飞机分割产生的“拖尾”现象，准确有效的分割出运动目标的阴影区域。再次，研究了基于加权形态学的泊位飞机的特征提取和基于神经网络的飞机识别方法。针对特征提取问题，提出利用加权形态学提取飞机的特征，解决了飞机机型匹配难题，形态学提取目标特征，同时去除了图像噪声，提高识别环节的准确性和效率，然后应用神经网络进行机型识别，取得了较好效果，实现低能见度下飞机机型识别，算法具有稳健性、准确性。最后，为了提高飞机跟踪的准确性，首先对均值漂移算法和传统粒子滤波算法的优缺点进行分析，在此基础上提出了递增自调整粒子滤波跟踪算法，采用递增自调整和位姿估计器使粒子向最优方向处移动，解决重采样过程中丧失多样性的问题，实现了采样粒子数量少，能够“智能”找到最优状态，仿真结果表明该算法具有良好的鲁棒性。仿真实验表明：所提出的算法对视觉飞机泊位引导系统的检测、机型识别和跟踪定位都有明显的提高，有一定的实用价值。更多还原

【Abstract】 The implementation of aircraft docking auto-guide to improve airport the level of informationand automation is essential, a method based on visual docking auto-guide because theinformation-rich, intuitive effects and low cost has been subject to the attention of scholars at homeand abroad. Image pre-processing and identification and tracking of the various aircraft in differentweather conditions restrict multifaceted uncertain factors, so the difficulty of visual docking guidanceand key technologies, thisarticle focuses on four aspects of work.Firstly, the conducting depth studies of image preprocessing algorithms of image enhancementand denoising in special weather conditions. In order to better adjust the contrast of image, thepiecewise linear transformation algorithm is proposed, which is simple, fast operation, to meet thereal-time requirements of the system. For fog weather, a novel improved edge detection method isproposed based on dark color priority and morphology defogging algorithm, which improves theclarity of the goal, preserves image detail edge to natural and realistic defogging, while meeting thereal-time and automatic requirements.Secondly, the segmentation of guided objects is researched. In edge detection, a new adaptiveweighted edge detection algorithm is proposed which takes into account the details of the edge andnoise removal. In the segmentation, taking into account the shadow of a moving object is marked forthe prospects to increase object tracking difficulty, matching the complexity of the algorithm as wellas posture evaluation algorithm, proposed a power transformation of the morphology of andborderless active contour model combined with the shadow segmentation algorithm, which not onlyfilters noise and preserves image details, improves the shadow segmentation precision, but alsoreduces the times of iteration. It also exploits the potential that active edgeless contour can be used inimages with complex backgrounds. Consequently preferably solve docking aircraft image objectpartition question.Thirdly, for the features extraction and recognition of aircraft objects, this article analyzes theinvariant moments of rotation, translation, scale invariance, a fast and effective identification methodis proposed which apply weighted morphological to be preprocessing simultaneous extraction of thecharacteristics of the aircraft, to solve the problem of aircraft models to match. Morphologicalfeatures in the feature extraction to remove image noise, improved recognition accuracy andefficiency, to achieve the robustness, the neural network model identification to solve the problem in the docking system, and achieved good results and accuracy of low visibility automatic dockingguided system.Finally, this paper analyzed the advantages and disadvantages of the MS and the traditionalparticle filter algorithms, an improved particle filter tracking algorithm is proposed to accurately trackthe docking aircraft Algorithm uses incremental self-adjustment and posture estimation and allowsparticles to move at the optimum direction, it solves the loss of diversity in the resampling process.The algorithm realized the small number of sample particles and capable of "smart" to find theoptimal state, the experiments show that the algorithm has good robustness.Simulation results show that: the proposed algorithms for the detection and model identificationand tracking of visual docking guidance system have significantly improved; there is a high value inengineering.更多还原