The number of the vehicles continues to increase from year to year and cause congestion especially when there is a long holiday. The cost of installing and maintaining the recording equipment and supporting infrastructure is quite extortionate. Therefore, the calculation of speed with a flexible and inexpensive cost requires drones. The purpose of this study is to monitor the flow of traffic in areas where there is no supporting infrastructure to install traffic cameras used to calculate the speed of the vehicles during long holidays. The camera shots taken from the drones are used to detect the types of passenger cars using library of You Only Look Once (YOLO) deep learning. The detected object is then tracked by using SORT (Simple Online and Realtime Tracking) to always detect the same object and calculate the speed of the vehicle when passing the RoI (Region of Interest). On the 5, 10, and 15 meters RoI, the lowest error values are 4.79%, 4.38%, and 2.96% respectively. Each of the results was obtained at a height of 10 meters. The proposed method can prove the error calculation of speed vehicle is small when the frame rate (fps) is high, altitude is low and the distance ROI is long.
Vehicle Speed Calculation from Drone Video Based on Deep Learning
2020-11-17
693419 byte
Conference paper
Electronic Resource
English