ABSTRACT
During the service period of the bridge, the vehicle load is the main variable load it bears. However, for dynamic traffic flow, due to its high degree of uncertainty and randomness, it is difficult to obtain actual traffic flow data with an efficient and accurate method. At present, commonly used methods for collecting vehicle driving characteristic data include: On-situ manual survey, Weigh In Motion, Induction coil system, Bridge Weigh In Motion, etc. In recent years, WIM system has become the main method to measure the actual traffic load of the bridge in service. However, the model obtained from WIM data can only accurately describe the traffic load characteristics of the cross section, and cannot obtain the traffic flow characteristics after the vehicle changes lanes. In this research, computer vision technology is used to realize the vehicle detection and vehicle tracking of the whole bridge, and then obtain the vehicle driving characteristics of the whole bridge. Specifically, our system can be installed on a long-span bridge, the cameras are arranged at a certain angle on the two pylons of bridge to obtain traffic flow images of the entire bridge deck. And based on YOLO v4 object detection and improved Kalman filter algorithm, the vehicles of the full bridge is recognized and tracked. In the end, the experiment on a suspension bridge proved that the system can realize the identification and tracking of vehicles on the full bridge. This research not only has practical value in studying the actual vehicle driving characteristics on the bridge, but more importantly, the method can be combined with the WIM system to obtain the time and space distribution information of the full bridge vehicle load.
