Research On Visibility Estimation Of Traffic Signs Based On Laser Scanning Point Cloud Data

Traffic signs,with different types and locations,are an integral part of transportation infrastructure,which are designed for giving traffic information for drivers and pedestrians and guiding traffic participants' behavior,resulting in smooth and safe traffic.As part of intelligent transportation system,the visibility of traffic signs is playing a significant role in road safety.Therefore,directive traffic signs should be salient and eye-catching.Nevertheless,there exists a number of factors that hinder its visibility in reality,among which obstacles of buildings and green belt caused by improper installation spot or height are the biggest.Drivers might go wrong if traffic signs were blocked,and traffic congestion and accidents might be caused because of drivers' slowing down to find direction.As a consequence,some are overly trimmed while others are ignored,which is neither efficient nor satisfactory.Thus,what is badly needed is an intuitive and accurate method that can evaluate the visibility of traffic signs.Nevertheless,traditional methods for the study of visibility of traffic signs were all based on computer vision,in which visibility of traffic signs were measured by features like contrast and texture between traffic signs and their surroundings in images and videos.These approaches don't take into consideration impacts exerted on traffic signs by factors like distance,visual angle,obstacles,etc,in three-dimensional space.Consequently,the thesis proposes a Laser-scanning-system-based method for the research on spatial visibility estimation of traffic signs.We can accurately measure the influence space feature can have on traffic signs,which is exactly the characteristic of this method.The research work explored in this dissertation are as follows:Firstly,we propose an effective algorithm for the extraction of traffic signs,we accomplish the detection task by making a combination of Reflected Intensity and Principal Component Analysis(PCA)to detect traffic signs in point clouds.To be specific,we first detect planar Point Clouds Cluster in the Point Clouds based on PCA,and then extract traffic signs in the Point Clouds Cluster using its high Reflected Intensity Feature.Secondly we propose spatial visibility field,intensity of spatial visibility field and visibility that measure objects' spatial visibility in 3D point clouds,and we define visibility as the effective image area of traffic sign in the retina.We make visibility as the standard of spatial visibility of traffic signs from different observing locations.Thirdly,we propose a framework of assessing spatial visibility of traffic signs.According to extracted traffic signs,we can calculate the distance between traffic signs,the visual angle and the size of the signs;and then we figure out the spatial visibility of the signs in question by adopting the quaternion method,alpha-shape algorithm and retina image formation principle;after that the shade is calculated by using X-ray transmission method and the viewpoint is figured out by spatial visibility field and shade?In this dissertation,we analyze and verify the proposed framework of spatial visibility of traffic signs through a series of related experiments,for instance,we collect Xiamen Island's road data by using Vehicle-Borne Laser Scanning System VMX-450 for experiment and validation.The results show that this framework of spatial visibility proposed in the paper can resolve the problem and achieve the research goal,it has important application value.