Research on Water Hazards Detection Method Based on A-MSRCR and Improved YOLO

The environmental perception capability of ground unmanned vehicles in unstructured environments in the wild is the premise for their path planning in passable areas, and obstacle detection is an essential part of environmental perception technology. In a complex environment, water hazards in the driving section of ground unmanned vehicles will pose a significant threat to the driving safety of vehicles. For example, water ingress into the interior of an unmanned ground vehicle will cause serious faults such as short circuits and sensor failure. Therefore, the research on the detection technology of water hazards in the unstructured environment in the field is of great significance and value. In the unstructured environment in the wild, the texture features of water hazards are not apparent, the features are easily blocked by the shade of trees or other shadows, the proportion in the camera field of view is relatively small, and the recognition rate is low. This paper proposes a fast detection method for water hazards in unstructured environments in the wild based on image enhancement and improved anchor frame YOLOv3.