Accident Detection from Drone Footage
R. Dev Abhishek, Vijaya Kumar Sundar · 2023
This research project aimed to address the issue of unreported accidents surpassing the time limit for reporting by leveraging drone surveillance in the context of smart cities and transportation. The project involved utilizing drone-captured images as input for an accident detection model, which incorporated a modified Intersection over Union (IoU) algorithm optimized with background subtraction to accommodate the resource limitations of an embedded environment. The implementation was conducted on a Raspberry Pi 4 B model, integrating a pre-trained model imported from a high-performance computer and layered with an anomaly detection algorithm. This algorithm not only indicated the presence of accidents in each frame but also provided a percentage estimation of accident presence along with identification of any additional anomalies suggestive of accidents occurring outside the current frame. To mitigate computational complexity, frame skipping was employed, skipping every other frame. The primary stage of the research yielded satisfactory results, demonstrating the model’s viability in contemporary society.