Developing Vehicular Traffic Datasets Using Drone images and AI Techniques

Developing Vehicular Traffic Datasets Using Drone images and AI Techniques Following the growth of cities, we are facing more traffic, pollution, and longer drives. Because urban road networks are complex and traffic conditions are changing, managing this traffic is difficult. Although there is the possibility of artificial intelligence as a solution, many current AI-based traffic systems are having trouble running consistently. The authors relied on data sets that were collected at fixed points, in stable environments, and updated infrequently. It is challenging for models to generalize to new cities, unusual incidents, or unexpected changes because such data sets cannot capture the diversity of real-world environments. This thesis describes a pipeline for using drone video and creating reliable urban data sets. As drones are easy to move and relocate, their wide coverage and adjustable viewing angles allow them to record traffic situation in different environments. After collecting data using drone an AI based pipeline that combines object detection, multi-object tracking, and speed estimation using calibration techniques processes the captured video. The study compares the advantages and disadvantages of different tracking frameworks by evaluating detection and tracking performance using different drone altitudes and viewing angles. It also looks at how accuracy changes with altitude and distance. Urban planning strategies, autonomous vehicle intelligence, safety evaluations, and the creation of more precise AI-based traffic models are all supported by the resulting datasets and processing pipeline. A strong calibration technique that reliably translates pixel displacements to actual speed measurements is one of the main contributions of this work. The resulting data sets are highly detailed and completely verified, offering a strong base for developing intelligent and flexible traffic control systems.