The idea of the proposed work can be summarized in Figure 1. First, we need to collect data on how traffic flows in the real world. This data helps us understand how diverse agent types move and interact with each other. The drone image dataset may be large in size and annotation of the complete dataset may not be easy. It is therefore proposed to use state-of-the-art vision algorithms to detect the different agent types along with their oriented bounding boxes. The agents’ oriented bounding boxes would be tracked with time to prepare a trajectory dataset of real life.
Accessibility Controls