Ad-hoc Aerial-view Vehicle Detection and Tracking for Real-Time Traffic Monitoring Using YOLOv7
Vishakha Singh, Phisan Kaewprapha, Choompol Boonmee · 2023
We have proposed an aerial view vehicle detection, classification, and tracking system using the You Look Only Once Version 7 (YOLOv7) machine learning model for automatic ad-hoc traffic statistical information gathering. The unmanned aerial vehicle (UAV) flies and gathers real-time visual data in areas where stationary cameras are unavailable such as rural roads or highway traffics. Eight different types of vehicles are detected and classified followed by tracking. Statistical data such as flow rate, direction, and car density are obtained afterward. The overall accuracy for the classification part is 93%, while the individual accuracies go as high as 99.5% for some classes.