Synchronized Multi Scale and Multi Sensor Traffic Data From Indian Urban Roads

With an alarming increase in congestion and pollution levels in the urban areas, especially in the cities, there is an imminent need for developing state-of-the-art transportation models to study alternate management strategies and solutions. Towards developing these, a comprehensive and high-quality traffic dataset is critical. Several transportation agencies across the globe have long identified a need for such datasets and initiated programs for data collection and management for research and engineering applications (e.g. Advanced Traffic Management System, Next Generation SIMulation, EUROPA).

Unfortunately in the Indian context, traffic dataset development activities have been piecemeal in nature since they are primarily generated as a part of specific studies. These datasets commonly suffer from three primary shortcomings: a) limited spatio-temporal extents of the data collected, b) a limited number of sources used and parameters observed, and c) limited to a single resolution of data (either macroscopic or microscopic). These shortcomings render the datasets inadequate and often unusable for multi-dimensional traffic applications. Furthermore, datasets available internationally do not adequately represent Indian traffic conditions.

To overcome these limitations, this research (for the first time in Indian conditions) proposes to develop a high fidelity traffic dataset for 25 kilometers of urban road system simultaneously utilizing a variety of data sources and measuring multiple parameters at both macroscopic and microscopic levels. The sensors proposed to be used in this study include WiFi sensors, Global Positioning System (GPS) sensors, video cameras, LIDARs, and drones. After obtaining permissions from the relevant authorities, the sensors will be field-deployed in the study area for data collection at both intersection-level and midblock-level. The macroscopic data such as travel time, flow, speed, density, classified vehicle count, geometry, origin-destination flow proportions, etc. will be collected. Microscopic data such as two-dimensional vehicular time-space trajectories will also be extracted for selected sections and intersections. GPS sensors will be installed on selected buses to characterize their movement and bus stops in the study area. The data from cameras and drone videos will be extracted using advanced machine learning and deep learning techniques currently being developed by faculty at RBC-DSAI. These techniques will be calibrated and validated with high-resolution LIDAR data. Moreover, the data will be processed through multiple filters to ensure high quality and consistency of traffic features across multiple data sources.

It is anticipated that the high fidelity traffic data generated in this study will capture the complex and finer interactions of various factors affecting driving behaviour that can be used to develop new and trustworthy traffic models. Moreover, it can help develop a range of new driving behaviour models to support the simulation of mixed traffic. The dataset is also expected to unveil the macroscopic and microscopic interactions across multimodal travelers, vehicles and highway systems, and traffic control devices. These will help to develop novel traffic management strategies. Finally, the dataset could also serve as a benchmarking system for researchers and engineers to evaluate various traffic management solutions.