New Methods for Accident Analysis Using Geographic Information System

本文共分为四个部分。首先，考虑到日本的现状，有必要从微观的角度进行考察。在本研究中，通过对事故发生地的微观考察，确定了道路的层次性、路边的违章停车、缩短视线距离的障碍物等因素。其次，结合交通事故、道路结构、土地利用等数据，构建交通事故分析GIS；在系统的构建过程中，创建了三种类型道路（全路、主干道和局部道路）的数据库进行分析，并利用这些数据库有效地找出了许多导致事故的因素。第三，建立统计模型，识别事故因素。回归模型是交通事故分析中常用的方法，但该方法是基于大样本的。在本研究中，由于泊松回归模型考虑了事故的罕见现象，所以建立了泊松回归模型。与回归模型相比，该模型具有更好的效果。最后，在基于调查数据的事故分析中，有趣的是，在人们认为危险的地区，实际发生的事故少于估计发生的事故。因此，开发新的模型来考虑这些心理因素。

In this paper, there are four parts. Firstly, considering the current circumstance in Japan, there is a need to investi-gate through the microscopic view. In this study, several factors, such as hierarchy of the road, illegal parking along the road and obstructions which make shorter the sight distance, are determined by the microscopic investigation of the places where the accidents occurred.Secondly, the GIS, integrated with traffic accident, road structure and land use data, is built for traffic accident analysis. In the process of buliding the system, databases for three types of road (all roads, trunk roads and local roads) are created for the analysis, and using these, many factors causing accidents are found out effectively.Thirdly, statistical models are also made to identify the factors of accident. The regression model is often used in traffic accident analysis, but, this method is based on large samples. In this study, Poisson regression model is built because it considers rare phenomena of accidents. Compared to the regression model, this model has better results.Finally, in the analysis of accidents based on survey data, it is interesting to note that in areas perceived by people as dangerous, the actual accident occurrence is less than the estimated occurrence. Therefore, new models are developed to take into account these psychological factors.