Abstract
Sixth-generation (6G) wireless technology has facilitated the rapid development of the Internet of Things (IoT), enabling various end devices to be deployed in applications such as wireless multimedia sensor networks. However, most end devices encounter difficulties when dealing a large amount of IoT video data due to their lack of computational resources for visual object tracking. Discriminative correlation filter (DCF)-based tracking approaches possess favorable properties for resource-constrained end devices, such as low computational costs and robustness to motion blur and illumination variations. Most current DCF trackers employ multiple features and the spatial-temporal scale space to estimate the target state, both of which may be suboptimal due to their fixed feature dimensions and dense scale intervals. In this paper, we present an adaptive mapped-feature and scale-interval method based on DCF to alleviate the problem of suboptimality. Specifically, we propose an adaptive mapped-feature response based on dimensionality reduction and histogram score maps to integrate multiple features and boost tracking effectiveness. Moreover, an adaptive temporal scale estimation method with sparse intervals is proposed to further improve tracking efficiency. Extensive experiments on the DTB70, UAV112, UAV123@10fps and UAVDT datasets demonstrate the superiority of our method, with a running speed of 41.3 FPS on a cheap CPU, compared to state-of-the-art trackers.