Development of collision prediction model under connected vehicle environment

Background

Safety is the primary concern in traffic engineering and improving safety is of paramount importance. With the advent of connected systems, proactive crash detection and mitigation of potential conflicts is becoming more feasible. These systems depend on accurate modelling of real-world vehicle behaviour to predict imminent risks. Traditionally, traffic models in this field focus on longitudinal movement using car-following models, which are designed for lane-based traffic. However, under urban traffic conditions, especially under multi-class lane-free (MCLF) conditions, traffic behaviour is also characterised by lateral movement, including frequent weaving, drifting, and turning. Moreover, classical models very rarely considered driver behaviour or vehicle dynamics in sufficient detail. In addition, there is a need to develop safety performance measures that can capture these characteristics. The current measures do not consider the above characteristics well, leading to errors in correctly identifying the safety critical situations. Thus, to enable more accurate and transferable safety solutions, there is a need for a hybrid modelling framework that integrates both longitudinal and lateral vehicle movements and quantify the driving behaviour well under diverse traffic conditions.

Research Gap

Current crash detection and safety warning models are based on models that are mostly developed considering the homogeneous and lane following traffic conditions. Many of the assumptions in them may not be valid or sufficient to capture the highly stochastic traffic conditions. Majority of them do not take the driver and vehicle characteristics in details and also consider only one-dimensional vehicle motion. Lateral vehicle dynamics remain underexplored, with few models that uses both data and physics aspects, particularly in the context of MCLF conditions. Furthermore, traditional car-following models overlook key behavioural factors such as reaction time, vehicle parameters, and decision-making variability. Thus, there is a need for a hybrid, longitudinal–lateral model that incorporates both vehicle dynamics and behavioural variability. Additionally, existing measures to quantify safety performance do not consider these characteristics. A lack of cross-context validation between structured and unstructured traffic conditions further limits the generalizability and real-world applicability of these solutions.

Aim and Objectives

The overall goal is to integrate longitudinal and lateral vehicle movements to develop a suitable safety performance measure through the connected vehicle data and use it to predict potential conflicts. The specific objectives are:

• Review existing safety performance measures, car-following and lateral movement models, and collect high-resolution data from homogeneous and MCLF traffic conditions for comparative analysis.

• Develop longitudinal vehicle movement models that incorporate vehicle dynamics and driver characteristics to realistically model the movements in the field.

• Develop a lateral movement model that can capture the nuances of vehicle interactions in the lateral direction, using a hybrid of physics based and data driven modelling.

• Derive safety performance measures that can capture the above modelled traffic behaviours and predict potential collisions under varying conditions.

• Implement and evaluate the developed framework under both homogeneous and MCLF conditions to assess effectiveness and real-world applicability of the models.