Traditional speed measurement methods, such as radar and laser speed guns, are expensive and require manual calibration, limiting their use. As a result, there is increasing interest in exploring vehicle speed measurement solutions based on traffic surveillance cameras, which offer a cost-effective, versatile, and potentially more reliable alternative. This approach allows for the use of existing traffic surveillance infrastructure.
In this thesis, we have developed an automatic vehicle speed measurement system using a single uncalibrated camera that operates without external scene information and in real time. Our solution is based on the YOLOv8 object detector for vehicle tracking and number plate detection. To calibrate the system, we use the UniDepth model for 3D scene estimation and the YOLOv8 model for license plate detection. Using the detected plates, we calibrate the depth model and use it to estimate the length of the road segment. Based on the estimated length of the road segment and the time taken by the vehicle to travel through the measurement zone, which is obtained from vehicle tracking, we calculate the speed of each vehicle.
|
