| description abstract | A disproportionate number of serious traffic accidents caused by lane offset occur at horizontal curvatures. Widely used methods for vehicle trajectory measurement, such as the vehicle position system and unmanned aerial vehicle (UAV), are unsuitable for wheel lane offset detection due to the resolution of data and shooting scope of cameras, respectively. To evaluate the lane offset risk of horizontal curvatures, automatic methods are proposed for horizontal alignment and vehicle trajectory measurement using a roadway inspection system (RIS), integrating a binocular stereo vision measurement system, initial measurement unit, and global positioning system (GPS). A mask region-based convolutional neural network (R-CNN) model is applied to detect lane markings. Based on binocular stereo vision technology, the lane offset of a vehicle on horizontal curvatures is measured continuously. To investigate the impacts of horizontal alignments on lane offset, inertial measurement unit (IMU) data and road scene images are applied for horizontal alignment measurement, including point of curve (PC) and point of tangent (PT) stations, curve length, curve radius, and turning direction. Four one-lane horizontal curvatures on highway ramps are selected as a test bed. Based on field data, the impact of horizontal alignments on lane offset is analyzed, and hazardous locations with lane offset risks are detected. This study can facilitate traffic safety analysis and the horizontal alignment design of roadway curvatures. | |
