基于小目标探测的高分辨率遥感影像交通参数提取研究

项目名称： 基于小目标探测的高分辨率遥感影像交通参数提取研究

项目编号： No.40801121

项目类型： 青年科学基金项目

立项/批准年度： 2009

项目学科： 交通运输

项目作者： 谭衢霖

作者单位： 北京交通大学

项目金额： 19万元

中文摘要： 民用m级及优于m级的航空航天高分辨率遥感影像车辆小目标的自动探测和分类，至今仍是一个挑战性的研究课题。项目研究提出了一种基于图像分割、综合上下文关系特征和模糊逻辑知识规则的面向对象遥感图像分析方法。首先，基于道路中心线生成道路掩膜，限制车辆探测在道路区域进行。其次，对生成的道路区域影像进行二次不同尺度的影像分割获得车道条带层和小目标对象基本层。然后，在小目标对象基本层构建面向对象的模糊分类器对该层对象进行车辆对象和非车辆对象分类；在已分类的小目标对象基本层上，融合相邻的同类对象生成车辆探测融合对象层，在该层上再分类车辆，获得完成车辆探测的影像。最后，基于分类识别车辆目标的平均长度和宽度，把车辆目标再分成小型、中型和大型车辆，并由计算机生成车辆目标总数和分类车辆数量。选择了多个代表性路段进行试验分析，结果表明：提出的方法在不同道路条件下均能有效地探测出高分辨率遥感影像中的道路车辆，具有良好的探测性能。由高分辨率遥感影像车辆小目标探测结果可直接获得道路车辆数量、类型、分布和路段/路线/路网交通车辆密度等交通参数，从而可为交通监测、规划和管理提供大范围的道路车辆交通场景信息。

中文关键词： 高分辨遥感影像；车辆探测；分割；面向对象；知识规则

英文摘要： High-resolution air-borne and space-borne sensors can now provide images with a resolution of one meter or decimeters, opening up for applications aiming at vehicle detection and traffic monitoring based on very-high-resolution remote sensing images. But the automatic detection and classification of small vehicle object in remote sensing imagery is still a challenging task. In the project, an object-oriented image analysis method has been developed to detect, classify and count road vehicles. The basic difference, especially when compared with previously developed pixel-based vehicle detection procedures, is that we don't process and analyze image pixels, but rather image objects that are extracted from image segmentation. A scheme was proposed to detect road vehicle objects based on fuzzy logic rule base. Firstly, a vector-generated road mask was used to constrain detection of vehicles to road region. Secondly, image segmentation algorithm was performed to form image objects in the preprocessing orthoimagery. Then, a fuzzy logic classifier, which based on a set of fuzzy logic rules defined by membership functions, was constructed to classify the extracted object regions into the vehicle and the non-vehicle regions by using the feature information of image objects. Finally, based on the calculated average length and width of vehicles, vehicle objects were classified into three categories, that is, small, medium and big. And the counts of the three vehicle classes were derived. A representative set of road segment images was selected from available images to test the proposed scheme. The extracted vehicle images were compared with the manually labelled vehicle images. Experimental results indicate that the proposed method has a good performance under varying conditions of road geometry, vehicle contrast, variability of pavement characteristics, and vehicle density. The detection rates of all test road-segments are high with very few false alarms. The scheme may find utility in generating initial estimates for vehicle counts. And further investigation and improvement could include improving automation by incorporating automated road extraction techniques before applying vehicle detection algorithms.

英文关键词： very high-resolution remote sensing image;vehicle detection;segmentation;object-oriented method; knowledge rules