A cyclist is a vulnerable road user whose interaction with the road infrastructure depends on several factors, including variable environmental conditions of lighting and meteorological road surface. This paper is concerned with nanoscopic crash modelling under the riskiest environmental conditions. There are very few works in the literature dealing with such modelling. An intelligent methodological framework consisting of the data collection unit and a knowledge processing unit (KPU) is proposed. In the knowledge processing unit, a combination of a) Statistical, b) Data learning and c) Casual inference methods are applied for investigating crashes on the study area of Tyne and Wear county in North-East of England. Three predictive nanoscopic road safety models are constructed (with 86% accuracy) using a) Spatial, b) Personal, and c) Infrastructure input variables. The importance of each of the identified input variable is estimated by deep learning and statistically validated through chi-square test and Cramer's V statistic. It is found that unsafeness of interaction between rider and infrastructure depends on lighting and road surface meteorological conditions. Different environmental conditions present a varying degree of risk to different types of infrastructure. The riskiest environment conditions are significantly affected by rider's gender and age, traffic flow regime, specific riding manoeuvre, and the road hierarchy difference. The increase in the number of variables, a rider encounters during his entire trip, imparts risky riding behaviour, affecting its safe interaction with the infrastructure. A novel infrastructure variable, i.e. `functional road hierarchy level and direction' introduced in this work, is found to be a critical road safety variable. A shift in road safety analysis towards nanoscopic modelling can help achieve zero-vision road traffic fatality. The study reinforces the need to plan and design infrastructure to move towards a more holistic approach while considering this vulnerable road user's limitations.
|Number of pages||12|
|Journal||IEEE Transactions on Intelligent Transportation Systems|
|Early online date||6 Aug 2021|
|Publication status||E-pub ahead of print - 6 Aug 2021|