A context-aware tracking method for aerial videos

Remote sensing target tracking in the aerial videos from aerial platforms is one of the research hotspots in visual tracking. In this paper, we propose a remote sensing target tracking method for aerial video based on a context-aware multi-domain convolutional neural network (CAMD). The process can be divided into two main stages: (1) in the design of the tracking network structure, we fuse multiple convolutional layers using residual connections to improve the effectiveness of regression learning. (2) in the “fuzzy interval”, a response-adaptive context-aware correlation filter (RA-CACF) module is introduced into our tracking network to boost the tracking performance. This method can greatly improve both the tracking efficiency and stability. We test the proposed method on the UAV123 datasets, and the experimental results demonstrate that our tracker can achieve high accuracy and efficiency results compared to state-of-the-art trackers.