Fast object detection from unmanned surface vehicles via objectness and saliency

Fast object detection is the most important part of the unmanned surface vehicles (USV) which make it possible for the USV to avoid the obstacle automatically and navigate autonomously. So, it is necessary to find a fast and accurate object detection method. In practice, the significant difficulty is that the environment is quite complicated which make the object uncertain. The obstacle may be a person, a sailing boat, a cruise ship, a floating speck or something else. At the same time, with the difference of size and angle of view, the same object may be quite different. The change of wave and light is another factor that affects the final detection results. Nowadays, there are many object detection methods, but they may be effective to special kinds of objects, furthermore the time cost is expensive, so these methods dont work very well for the USV. In this paper, we propose a new method to solve these problems. Firstly, we use the objectness to get the object proposals. The proposals may have many false alarms. Secondly, though the environment is various, the obstacle around the USV is very distinctive, so we apply the idea of saliency to calculate the salient map. Finally, fusing the result of objectness and saliency, compute the salient density for each object proposals and remove the proposals with false alarms. We test our algorithm on several challenging dataset which is collected in marine environments, the results show that our algorithm has high accuracy and low false alarm rate.