In order to ensure that ships in inland waterways can realize safe and efficient autonomous navigation, this paper proposes a method for real-time waterway detection using MonoLayout algorithm, which predicts the Bird’s Eye View (BEV) in front of the ship using the First Person View (FPV) image as input. In order to improve the accuracy and generalization of the model, two datasets are produced for training, namely, the dataset containing only the real scene (RealBEV) and the dataset with 7:1 mixing of the Unity3D simulated scene and the real scene (MixBEV). To validate the data and recognition effects of comparing the two datasets, multiple experiments are conducted. The experiments show that the model trained with MixBEV has an average intersection and merger ratio (mIoU) of 88.02%, an average pixel accuracy (mAP) of 89.49%, and a single-image prediction time of 24.06 ms for channel detection. mIoU and mAP are improved by 5.58% and 3.19%, respectively, compared to the single RealBEV data, and for different ships in the same scene, The generalizability is well improved. It also shows that MixBEV can better balance the accuracy of recognition and real-time recognition when detecting inland waterways.