基于Mean-Shift理论的智能车辆道路检测方法
本文就Mean-Shift算法的道路检测系统进行研究，主要工作包括两部分：(1)研究并实现了基于Mean-Shift理论的图像分割算法用于透视道路图像的分割；(2)研究并实现了基于Mean-Shift理论的区域跟踪算法用于对逆透视图像中道路区域的跟踪。
具体为，将车载摄像头拍摄得的实时图像通过逆投影映射，将实时图像变换为逆透视投影映射图（Ipm图），对远处的干扰进行模糊化处理，以减少外景对于跟踪算法的干扰问题。针对近处需要跟踪的路面情况进行细化，使跟踪过程变得更加准确，使算法性能更优。针对车载摄像头拍摄的实时道路图像，道路的转向问题，采用了对图像进行分别左右旋转一定度数，再进行跟踪，分别得出与上一帧图像跟踪目标比较的相似度，将相似度进行比较，从而得出一个最佳的跟踪角度。又以该角度为基准对下一帧图像进行旋转跟踪过程，对图像序列进行迭代直到跟踪算法结束。跟踪过程结束后，将跟踪框反变换到分割图像上，在被跟踪框覆盖到的区域中，跟踪框内像素点个数占该区域像素点个数超过一定百分比，就认为真实道路区域包含该分割区域，从而得出真实道路区域。关键词 均值漂移 道路检测 图像分割 图像区域跟踪 逆透视投影
毕业论文设计说明书外文摘要
Title Intelligent vehicle road detection method based on Mean-Shift theory
Abstract
The road detection system is an important part of the unmanned vehicle driving system， and the progress of the road detection technology can produce a huge power for the unmanned vehicle driving system。。 Road detection technique is for the vehicle mounted camera captured image detection process， get the image information and unmanned vehicle to road information and feedback to the unmanned vehicle driving system。 Therefore， the study of an accurate road detection algorithm is a important research study meaning of。
The mean shift algorithm of road detection system was studied， the main work includes two parts: (1) is studied and realized based on the theory of mean shift image segmentation algorithm for perspective road image segmentation; (2) study and realize the used mean shift theory of region tracking algorithm of inverse perspective road image region tracking based on。
Specifically， the vehicle mounted camera shooting the real-time image by inverse projection mapping， real time image transform for inverse perspective projection mapping (IPM)， interference in the distance of fuzzy processing， to reduce the location for the disturbance of the tracking algorithm。 According to the road conditions need to refine close tracking， make the tracking process more accurate， the method has better performance。 For a vehicle mounted camera real-time road image， road to the problem， the image respectively about is rotated to a certain degree， to carry on the track， were obtained with a frame image tracking goal to compare the similarity， the similarity comparison， to arrive at an optimal angle tracking。 Taking the angle as the reference to the next frame image， the image sequence is iterated until the end of the algorithm is finished。。 After the end of the tracking process， will track frame inverse transform to image segmentation， the tracking frame including area， within the frame number of pixels for the region pixel number exceeds a certain percentage of track， that real road region contains the segmented regions， so as to obtain the real road area。
上一篇：基于谱聚类理论的场景分类研究
下一篇：面部轮廓识别与机器人素描技术研究