Abstract:
Traffic congestion has a significant negative impact on the environment
and on people, and one of the most important reasons is poor control of
traffic signals. Vehicle counting in crowded intersections is a crucial element
of any traffic optimization solution. In this work, we propose a computer
vision based vehicle counting method. We adopts YOLO-v3 as vehicle de tector, we added a series of post-processing mechanisms to achieve robust
vehicle counting. We created framework to asemi-utomatically generate a
dedecated dataset for vehicle counting, this dataset can be later used as
training data for new maching learning based solutions.
