Abstract
The aim of this study is to test the entire framework on an AI demo board and explore the problem of how to automatically and effectively detect and analyze traffic accidents captured by surveillance cameras. First, the motion interaction field (MIF) method, which can detect collisions in videos, is used to locate the location of the damaged vehicle based on the interactions of various moving objects. Then, the YOLO v3 model is used to locate the location of the destroyed vehicle. Using a hierarchical clustering method, the vehicle trajectory before the collision is restored and the related trajectory is reconstructed. Finally, a perspective transformation is used to project the trajectory into a vertical view, allowing traffic police officers to make better decisions. The vehicle speed is estimated using the Undistorted Finite Impulse Response (UFIR) method, which does not require statistical information on external noise.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call