都市計画研究室

交通・都市環境計画研究室

都市・地域デザイン研究室





 

2019/11/30-12/02 第60回土木計画学研究発表会・秋大会で発表しました

2019/09/19  リーズにてセミナーを行いました.

2019/09/09-11  Colomboにて学会発表をしました.

 

第60回土木計画学研究発表会・秋大会で発表しました

 
11/30-12/2に富山大学にて愛媛大学から4名発表を行いました.   

深層学習による首都高速道路の事故発生予測モデル

森本裕治

時間的・空間的に広範囲の感知器情報を入力に用いて,予測時点直後の特定道路区間における事故の起こりやすさを予測するConvolutional Neural Networkモデルの構築を試みる.具体的には,首都高速道路の複数路線を対象として,路線上に存在する全感知器による過去1時間の観測情報を入力として,予測時点から2時間先までの間に事故が発生する確率を予測するCNNモデルを構築する.続いて,過去9年間のデータを用いてモデル学習を行い,構築される事故発生予測モデルの検証を行う.検証の結果,多数の織り込み区間が存在する路線に於いて事故リスク予測精度が低下するとともに,一定のサンプルが獲得されれば高精度で事故発生確率を予測することができるとの結果が得られた.


4回生ながら堂々と立派に発表しました!!

   

自動車交通事故に対する恐怖感に着目した経路選択意識構造の分析

倉内慎也

交通事故統計によれば,高規格な道路ほど死傷事故率が低いことが明らかとなっている.一方,高速道 路の利用を対象とした著者らの先行研究では,高速道路の事故リスクを過大に知覚しているドライバーが 一定割合存在すると共に,そのような知覚バイアスが運転に対する恐怖感を介して高速道路の利用を妨げ る一因となっていることが確認されている.しかしながら,一般道路における事故をめぐる意識について は,これまでほとんど研究がなされていない.そこで本研究では,一般道路を対象に,交通事故に対する 恐怖感や幹線道路の利用をめぐる意識構造を分析し,幹線道路の利用促進に向けた効果的な事故リスクコ ミュニケーションについて検討した結果を報告する.



一般街路における舗装材質を考慮した事故リスク分析

坪田隆宏

本研究では,一般街路を対象に舗装材料の違いが事故リスクに与える影響を明らかにする.道路舗装は 車両の走行安全性や快適性に直接的にかかわる交通基盤であることから,その性能を最大限に発揮するべく,環境や道路線形等を 考慮した多様な舗装材質が使用されている.舗装材質の性能のうち,走行安全性にかかわるものの一つに滑り抵抗性があり, 舗装材質別の同性能は値は実験環境下において明らかにはなっているものの,実共用下における舗装材質と走行安全性に の関係を分析した事例はない.そこで,本研究では,舗装材質の違いが事故リスクに与える影響を明らかにする. 具体的には,舗装材質別に道路構造を考慮した事故リスク値の算出を行った.北海道における直轄国道全路線を対象とした 集計分析の結果,滑り抵抗性が低いとされる舗装材質区間において事故リスクが高まる傾向にあること,また, 舗装材質別の事故リスクは道路構造による影響を受けることが明らかとなった.今後は舗装材質を考慮した事故リスク推定モデル の構築に取り組む.

   


情報の提示方法を考慮した事故リスク情報提供効果分析

吉井稔雄

潜在的な交通事故の危険性に関する情報,すなわち“事故発生リスク”情報を提供することにより,より安全性の高い経路へ需要をシフトすることによって,個々のトリップの安全性が向上し,事故削減につながることが期待される.道路利用者に対して経路誘導へのインセンティブを与える可能性を有する事故リスク指標として,事故発生リスク情報に加え,高速道路走行時に事故現場に遭遇する可能性を表す“事故遭遇リスク”,高速道路走行時に失う可能性のある損失額である“事故損失リスク”の2指標が提案されている.本研究では,これら質の異なる事故リスク指標を道路利用者に提供する場合に期待される経路誘導効果と事故減少効果の評価を実施する.具体的には,経路選択を有するNEXCO 西日本のネットワークを対象とし,事故リスク情報を考慮した経路選択モデルを用いて,各種事故リスク指標を提供した場合の道路利用者の経路選択率への影響を分析し,事故リスク情報提供による事故減少効果を定量的に示した.

   
   

Leedsにてセミナーを行いました


9/15-9/23にてイギリスLeeds大学にてセミナーを実施しました.





Effect of traffic accidents on regional road networks

Ms. Rui Okuhara, Ehime University

Once a traffic accident occurs, the loss due to the accident is generated not only in vicinity of the accident cite but also in surrounding wide-area road network. This study establishes an evaluation method of the impact of a traffic accident on the traffic flow of a regional road network. First, an evaluation method of representative velocity of the network traffic flow is developed by using prove trajectories. Then, it is applied to the Matsuyama regional road network, and the relationship between the velocity and occurrence of a traffic accident is investigated using analysis of variance. The results show that a traffic accident has a significant effect on the represented velocity of the network traffic flow.






A Study on the Characteristics of MFD determined by Probe Data

Mr. Shingo Hayashi, Ehime University

Mr. Hiroto Nakatou, Ehime University

This paper investigates the characteristics of the macroscopic traffic states represented by the Macroscopic Fundamental Diagram (MFD). Factors that have influence on the MFD shapes have been intensively reported. However, comprehensive study on the MFD shapes from various types of cities has not been investigated. This study investigates the characteristics of the MFD shapes from 47 cities in Japan. The MFDs are defined by the nationwide probe vehicle system in Japan. The analysis is performed in the 47 prefectural capitals. The MFDs are obtained by performing a piecewise linear regression of the area states consisting of standardized traffic flow and density. The results of the k-means method identified 4 typical shapes of MFDs, indicating that the MFD shapes could be affected by types of cities. Future research attempt includes understanding how the traffic characteristics and road network structures in the 47 prefectural capitals are affecting the shape of their MFDs.






Prediction of Dynamic Traffic Accident Risk on Tokyo Metropolitan Expressway by Artificial Intelligence

Mr. Yuji Morimoto, Ehime University Mr. Mamoru Shinmizu, Ehime University

This study develops an on-line AI model for predicting the likelihood of the occurrence of an accident. The AI model with Deep Neural Networks can predict the likelihood on the next 30 minutes. The input data consists of time-series traffic detector data, precipitation and others. The model outputs the probability of an accident occurring during the next 30 minutes. Then, the AI model is applied to the Tokyo Metropolitan Expressway network. Machine learning task is carried out on the training set of time-series data in 6 years. Then the performance of the training model is evaluated on the test set, which consists of them in a month outside the specific period above. As a result, the accident risks on the next 30 minutes have been well predicted in a high accuracy.






Advanced Traffic Management for Smart Mobility

Effect of a Safety Route Guidance System on Network Traffic Safety Prof. Toshio Yoshii, Ehime University

his study evaluates the effect of a safety route guidance system on network traffic safety. First, dynamic accident risk on each road link is understood by analyzing the historical data. Then, an accident risk estimation model is constructed, which can evaluate the effect of weather and traffic states in addition to static ones, such as gradient, curvature and others. Second, a safety route guidance system, which provides the lower accident risk route to drivers, is established. Then, the demonstration test is carried out where the lower accident risk route is provided to actual drivers through the navigation system (see Fig. below). Finally, the effect of the safety route guidance system on network traffic safety is evaluated by a simulation analysis. As a result, 70% of the drivers chose the lower accident risk route in the demonstration. Also, it is shown that 12% of accidents on the network can be reduced by introducing the safety route guidance system.



Fig. Display of the safety route guidance system

Dr. Toshio Yoshii is a professor of Graduate School of Science and Engineering, Ehime University, Japan. He graduated from The University of Tokyo in 1992, and got Ph.D. in 1999 supervised by Dr. Masao Kuwahara. Then he has made contributions in as diverse transportation research topics as: traffic flow modeling, dynamic assignment, traffic safety, travel behavior modeling and others.







Colomboにて学会発表をしました.

9/9-12にてColomboにて学会発表を行いました.


A Study on the Characteristics of MFD determined by Probe Data

Stephanie Gituru, Ehime University

Traffic congestion, one of the main causes of various social and economic problems has become a persistent issue around the world. The Macroscopic Fundamental Diagram (MFD) aids in understanding the traffic conditions of an urban network; which is paramount in the management of this challenge. This study investigates the characteristics of the MFD shapes from 47 cities in Japan. The MFDs are defined by the nationwide probe vehicle system in Japan. The analysis is performed in the 47 prefectural capitals. The ETC system has gained popularity in the recent years, subsequently leading to an increase in the number of ETC equipped vehicles. For this reason, the area states were standardized to take care of the growth in sample size. The MFDs are obtained by performing a piecewise linear regression of the area states consisting of standardized traffic flow and density. K-means method was used to cluster the MFDs. The results of the k-means identified 4 typical shapes of MFDs, indicating that the MFD shapes could be affected by types of cities. Future research attempt includes understanding how the traffic characteristics and road network structures in the 47 prefectural capitals are affecting the shape of their MFDs.


Analysis of the Safety Performance of Drainage Pavement focusing on Pavement Age

Celso FERNANDO, Toshio YOSHII, Takahiro TSUBOTA, Hirotoshi SHIRAYANAGI

Abstract: This study analyzes the relationship between accident risk and the age of drainage pavement also known as porous asphalt. The analysis is based on empirical data from an urban expressway in Japan. Drainage pavement has been the most common type of road pavement in urban expressways due to its permeability function, which provides a safer driving environment in wet weather. However, this function decreases over time, and timely maintenance is necessary to ensure road safety. It is therefore essential to understand the change in the safety performance over the service life of drainage pavements. This study defines safety performance as the number of accidents per unit of vehicle kilometer traveled, considering factors such as pavement age, weather, and road geometry. Statistical analysis based on the Poisson regression model revealed that, under wet condition, pavement age has a positive effect on accident risk mainly on curve segments.






Effect of Attentional Disengagement on Driver Inattention While Driving on Expressway

Hirotoshi SHIRAYANAGI, Shinya KURAUCHI, Takahiro TSUBOTA and Toshio YOSHII

This study aims to demonstrate that driver’s detection of a fallen object on an expressway can be improved by attentional disengagement. Driver’s detection of a target often delays due to vigilance decrement, which is deterioration in the ability to keep attention to a target over a long period of time. Previous studies have suggested that instantaneous attentional disengagement from a target relieves vigilance decrement. It is considered that driver’s attention tends to be located on a particular point in front for a long time which could cause vigilance decrement resulting in the delay in detection of a fallen object on an expressway. This study conducts a selective adaptation task to show that attentional disengagement relieves vigilance decrement during driving on an expressway. The response time of the disengaged condition was observed to be shorter than the engaged condition. This result suggested that attentional disengagement led to improve the driver’s detection of a fallen object during driving on an expressway.





@2019 交通・都市環境計画研究室/都市・地域デザイン研究室