基于深度学习的车辆检测与消除方法在倾斜摄影自动三维建模中的应用

发布时间：2018-01-19 21:50

本文关键词： 城市三维模型深度学习车辆探测目标消除摄影测量　出处：《武汉大学》2017年硕士论文　论文类型：学位论文

【摘要】：高精度三维城市模型是智慧城市的基础数据,是各种城市应用所需要的基础信息的载体。近年来,随着摄影测量粤计算机视觉技术的突破,利用倾斜摄影采集的多角度图像进行城市三维模型重建已经日益被广泛应用。在三维自动建模过程中,模型质量除了会受到分辨率、重叠图等各种因素影响之外,道路上的移动对象对三维模型质量的影响也非常大,常常造成道路模型的严重扭曲以及纹理的错位,其中,道路上的移动的车辆造成的模型变形尤其严重。如何自动的消除移动测量对建模质量的影响是一个非常重要而且具有实际价值的问题。本文针对此问题提出一种图像预处理方法,采用深度学习技术,从实景图像中识别运动状态的汽车,在图像中擦除检测到的移动车辆,并用周围的道路图像填补车辆擦除形成的空洞。此方法为空中三角测量及三维建模前的图像分离处理,旨在消除图像中由运动物体带来的延迟效果,并最大限度的保持其表达真实性和准确性。本研究从计算机视觉和图像理解的角度出发,解决两个问题:目标探测和目标移除。首先,设计和应用基于深度学习的目标分类方法检测出倾斜影像中移动状态的汽车,然后使用纹理合成及图像修复技术将其移除,最终将结果图像用于城市三维模型的构建。实验表明,本文所提出的方法能够有效提升由车辆移动而造成的路面模型扭曲以及纹理错位。
[Abstract]:High-precision three-dimensional city model is the basic data of intelligent city and the carrier of basic information needed for various urban applications. In recent years, with the breakthrough of computer vision technology of photogrammetry in Guangdong. In the process of 3D automatic modeling, the quality of the model is not only the resolution, but also the resolution in the process of 3D automatic modeling by using the multi-angle images collected by tilt photography to reconstruct the urban 3D model. In addition to the influence of overlapped images and other factors, moving objects on the road also have a great impact on the quality of 3D models, which often cause serious distortion and texture dislocation of road models. The model deformation caused by moving vehicles on the road is especially serious. How to automatically eliminate the influence of mobile measurement on modeling quality is a very important and practical problem. In this paper, a new method is proposed to solve this problem. Image preprocessing method. Using the depth learning technology, the moving vehicle is identified from the real scene image, and the detected moving vehicle is erased in the image. The surrounding road image is used to fill the void formed by vehicle erasure. This method is used to separate the images before aerial triangulation and 3D modeling in order to eliminate the delayed effect caused by moving objects in the images. And to maximize the authenticity and accuracy of its expression. This study from the perspective of computer vision and image understanding to solve two problems: target detection and target removal. The object classification method based on depth learning is designed and applied to detect the moving vehicle in the tilt image, and then it is removed by texture synthesis and image restoration techniques. The experimental results show that the proposed method can effectively improve the distortion and texture dislocation of the road surface caused by the movement of the vehicle.
【学位授予单位】：武汉大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;P23

【相似文献】