Executive Summary : | The project aims to develop an intelligent pedestrian crossing framework that facilitates non-verbal communication between pedestrians and vehicles by warning them in advance. The framework will automatically detect and analyze near-miss events at pedestrian crossings, identifying the cause(s) behind their occurrences. The project also aims to model and reliably predict the future state of pedestrians and yielding intentions of approaching vehicles using their characteristics, past trajectory, 2-D skeletal pose of pedestrians, and surrounding context. The next hypothesis is to model the occurrence of severe traffic conflicts using a deep learning framework, able to anticipate safety-critical events before a reasonable period of time and warn road users to avoid collisions. The trained deep learning models will be tested in a simulated environment, and the most reliable one will be deployed at the site. The expected result of this project is the deployment of deep learning models trained at pedestrian crossings to monitor aberrant driver and pedestrian behavior, predict future behavior, and warn crossing pedestrians and approaching vehicles. These models can be deployed through infrastructure-to-vehicle (I2V) connectivity, allowing for reliable transmission of data fully adapted to pedestrian crossing or driver yielding behavior or advanced driver assistance systems (ADAS). The intelligent framework developed using data from different parts of India will be immensely useful for modern smart cities, ministries, police departments, NGOs, and other organizations. The project's outcomes will have commercial importance, with the research findings being converted into products and the idea and product obtained being patented. |