A Real-time Horizon Detection Method Based on Confidence Map in Maritime Scenarios

Horizon detection is of great significance to unmanned surface vehicle in maritime scenarios. Due to complex lighting and changeable ocean conditions, horizon detection is still a challenging task nowadays. In order to solve the problems, we propose a new dataset called Taihu Horizon Dataset and a new approach based on confidence map for horizon detection task. Our method leverages the power of deep convolutional neural network to predict the the probability of the class of each pixel. Then non-maxima suppression are performed on the confidence map to obtain the fitted curve. We conduct experiments on Taihu Horizon Dataset and Singapore Maritime Dataset. Comparison of experimental results turns out that our approach has excellent performance and achieves a good balance between accuracy and real-time performance.