ITD: Indian Traffic Dataset for Intelligent Transportation Systems

Efficient transportation has always been the driving force behind the pace of the nation's progress. In recent years, several large-scale driving datasets capable of handling a myriad of tasks related to Intelligent Transportation Systems (ITS) have become a necessity. However, no existing dataset (a) caters to the diverse vehicle types plying on traffic streams in developing economies such as India (b) facilitates the real-world applications of the videos/ images from a static camera set-up. To address these challenges, an Indian Traffic Dataset (ITD) is developed in the present study by capturing diverse traffic vehicle mixes from different regions in India. Additionally, the set of vehicle classes is labeled as per the Indian Highway Capacity Manual (INDO-HCM), which makes it unique from the existing datasets. This classification is likely to be suitable for other South and South East Asian countries due to the abundance of auto-rickshaws, electric-rickshaws, motorized two-wheelers, etc. ITD comprises 17,666 annotated images, of which 7,778 are novel images; these images have nine classes aggregating close to 341,000 annotated objects. The proposed dataset has shown no significant difference between observed and predicted traffic volume count at a 5% level of significance. The dataset possesses diverse traffic conditions ranging from structured to unstructured road environments to enable heterogeneous task handling, such as vehicle counting and traffic flow analysis, and can be extended to trajectory analysis, safety assessment, traffic violation detection, etc.