测绘学报 ›› 2015, Vol. 44 ›› Issue (1): 39-45.doi: 10.11947/j.AGCS.2015.20130358

• 摄影测量学与遥感 • 上一篇    下一篇

应用光照模型的交通车辆识别定位的高分辨率遥感方法

曹天扬1,2, 申莉1,2   

  1. 1. 北京遥感设备研究所, 北京 100039;
    2. 北京航天福道高技术股份有限公司, 北京 100195
  • 收稿日期:2013-12-11 修回日期:2014-04-16 出版日期:2015-01-20 发布日期:2015-01-22
  • 作者简介:曹天扬(1984-), 男, 博士, 研究方向为信号处理、图像处理、多传感器信息融合等. E-mail: cao_tian_yang@sina.cn

High Resolution RS Image Processing Method for Vehicles Recognition and Location According to the Illumination Model

CAO Tianyang1,2, SHEN Li1,2   

  1. 1. Beijing Institute of Remote Sensing Equipment, Beijing 100039, China;
    2. Beijing Aerospace Fudao High-Tech Co Ltd, Beijing 100195, China
  • Received:2013-12-11 Revised:2014-04-16 Online:2015-01-20 Published:2015-01-22

摘要: 针对识别和定位路面上每辆汽车这一交通遥感图像处理的核心环节,提出一种解决方法.应用光照模型推导出路面、车辆(深浅两类)、汽车阴影在全色遥感影像中的亮度差异与亮度变化特征.以亮度差异为基础建立了能够将车辆区域从路面遥感图像中分割出来的图层分离算法.针对密集行驶的汽车因阴影相互覆盖而容易被误识别为一辆大型车的问题以及浅色车因深色车窗造成的识别结果割裂问题,利用亮度变化特征以及阴影、车窗与汽车的位置关系设计了车辆区域内的阴影和车窗干扰消除算法,通过闭运算实现了遥感图像中的汽车识别与定位.选用10幅交通遥感图像进行了测试,对浅色车的识别率大于92%,对深色车的识别率大于87%.

关键词: 交通遥感信息处理, 车辆识别定位, Phong光照模型, 阴影滤除, 车窗干扰

Abstract: For the issue of vehicles recognition and location, which is the key of traffic RS information processing, an image processing method is presented. Based on the Phong model, features of luminance difference and luminance variance of common objects, including vehicles (dark and bright) and their shadows and road surface, are extracted. By taking advantage of the features of luminance difference, the vehicle regions, containing vehicle and its shadow, are extracted through layer separation firstly. And then, in order to suppress interference caused by vehicle shadow which is easy to cause the connection between abreast running vehicles, as well as suppress interference caused by vehicle window which might disturb the bright vehicle identification, an algorithm is designed according to the features of luminance variance and the position relationship between vehicle and its shadow and window. Through morphological method, the vehicles of different colors are extracted and located on the RS-image. This method had been tested, and more than 92% bright vehicles and 87% dark vehicles are detected.

Key words: traffic RS information processing, vehicle identification and location, phong illumination model, shadow suppress, windows interference

中图分类号: