A Highway Distance Posts Detection Method Based on S-YOLOv3

In GPS denial environments, the geographic information of the location of distance posts could be obtained through detecting these posts in images, and could be used for vehicle positioning. Resolution of vehicle-mounted camera imaging will be affected by complex environments, changing imaging scales and angles that bring difficulties to the detection. Although YOLOv3 is effective to solve this question, its architecture is too complex to calculate. Aiming at the requirements that there are only two types of distance posts to detect and fast operation speed in our task, we put forward a new form of YOLOv3, named S-YOLOv3 (simplified YOLOv3). Through video images acquired from the Daheng mercury camera mounted in the vehicle, we completed training and testing of the network. According to the results, our method has better detection robustness, whose mAP(mean Average Precision) reaches 85% and detection speed reaches 28FPS.