In India, different vehicles have different ways of handling the congestion. Bikes can swerve in and out between vehicles and reach the front of the queue at the signal. Auto rickshaws, too, have this uncanny ability to sneak between other vehicles to quickly get through the junction. However, large vehicles like SUVs, buses, and trucks don’t have much freedom in heavy traffic. However, when there are sufficient oversized vehicles, they dominate the roads since the bikers feel vulnerable during those conditions. All these practical nuances need to be considered when evaluating traffic conditions and implementing traffic management strategies.
However, the state-of-the-art models and theories are based on homogeneous conditions. In other words, they consider only one class of vehicle: car. Thus, the popular characterisation of traffic state is based on speed, flow, density, queue length, etc., which are inadequate to represent the mixed traffic conditions commonly found in India since they do not consider multiple vehicle classes and their interactions.
In this study, the authors, Ms. Abirami Krishna Ashok and Prof. Bhargava Rama Chilukuri from the Department of Civil Engineering, Indian Institute of Technology (IIT) Madras, Chennai, India, have characterized, estimated, and predicted traffic states for mixed traffic conditions.
A speed-based characterisation of traffic state and a joint class-wise speed prediction methodology have been proposed in this study. All types of vehicles commonly found in India are considered – two-wheelers, three-wheelers, cars, light commercial vehicles, and heavy vehicles (including buses). The novelty of this method is the use of class-agnostic speed data (speeds are known, but the vehicle type is not known, generally obtained from sensors such as Wi-Fi Media access control Sensors and Bluetooth sensors) that was used to characterize the traffic state. The travel time data of multiple vehicles collected during a five-minute period was pre-processed into binned data. Data-driven state characterization and estimation methodologies were proposed using binned data. Classical traffic flow principles and models are used to estimate class-wise speeds under varying traffic conditions. The characterization and estimation methodologies endogenously account for bounded accelerations of vehicles and driving behavior under different congestion levels and traffic compositions.
A joint model approach was proposed in this study that simultaneously estimates the class-wise speeds from the traffic state. A parametric model (Logistic Regression) and a non-parametric (Multi-Layer Perceptron) model were chosen to predict the traffic state and benchmark the performance of the joint prediction model with the marginal (or disaggregate) prediction models from the literature. The results show that the proposed models are superior in accuracy, but more importantly, they preserve the speed-ordering of the vehicles. This approach allows us to estimate and predict class-wise speeds more accurately in mixed traffic conditions observed in Indian cities.
This research opens doors for a better understanding of multi-class behavior in mixed traffic conditions, thus enabling the development of a new class of traffic management strategies based on class-wise behavior. Based on these behaviours (of two-wheelers, three-wheelers, cars, trucks, etc.) and their respective speed profiles, traffic managers can design interventions that specifically target each vehicle class through strategies such as dedicated lanes or scheduled restrictions of various vehicle classes to alleviate congestion and reduce emissions.
Dr. C. Mallikarjuna, from the Civil Engineering Department, Indian Institute of Technology (IIT) Guwahati, Guwahati, India, acknowledged the importance of this paper with the following comments: “For many decades characterizing the heterogeneous no-lane disciplined traffic stream has been a challenging task for the researchers. Traffic stream characterization is critical for effective traffic management as well as for developing tools for intelligent transportation systems. This study gave insights into the fundamental understanding of class-wise speed-based traffic state prediction that is critical for traffic stream management. The proposed models are expected to be instrumental in advancing traffic management at the network level and will likely provide a foundation for future research in this area.”
Article by Akshay Anantharaman
Click here for the original link to the paper
