An Integrated Detection and Tracking System Applied to Unmanned Surface Vehicles

Perceiving surrounding environment is essential in the navigation, obstacle avoidance and other tasks of Unmanned Surface Vehicles (USV), in which object detection and tracking play vital roles. However, it is susceptible to the influence of occlusions, scale variations and severe shaking, in the meantime requiring higher speed. We introduce an integrated detection and tracking system applied to USV, which combines the tracking algorithm based on improved adaptive scale correlation filter with the detection algorithm based on deep learning. After tracking the fixed number of frames, we perform the re-detection mechanism and utilize the bounding boxes selection strategy to determine subsequent tracking boxes. We conduct experiments in practice and the results demonstrate that the system can continuously detect and track objects on the water at real time with the speed of 15FPS and the accuracy of 0.87, which illustrates robustness is improved on the condition that time consumption of detection and tracking is reduced.