Qiqi Chen

Visual tracking on unmanned aerial vehicle (UAV) platforms is a fundamental and crucial visual task. Compared to conventional tracking tasks, aerial tracking faces specific challenging scenarios due to its unique perspective. Although existing aerial trackers have demonstrated promising performance, they remain limited in capturing spatial-channel saliency across branches and effectively utilizing historical information. To address these issues, this letter proposes a spatial-channel saliency and temporal fusion network (SiamSCT), which aims to enhance feature representation for efficient and accurate aerial tracking. Specifically, SiamSCT introduces a weight-shared spatial saliency block to strengthen the spatial feature representation across the tracking network’s branches. In addition, a light channel aware module is designed to facilitate deep feature interaction across branches at the channel level, further enhancing feature discriminability. Finally, using a historical similarity response fusion strategy, SiamSCT achieves more stable and reliable tracking responses, effectively tackling complex aerial scenarios. Extensive experiments on several authoritative aerial tracking datasets demonstrate that SiamSCT outperforms state-of-the-art trackers. Furthermore, SiamSCT achieves a tracking speed of 133 FPS on NVIDIA RTX 3060Ti, proving its excellent performance in real-time.

In recent years, many studies have used Siamese networks (SNs) for UAV tracking. However, there are two problems with SNs for UAV tracking. Firstly, the information sources of the SNs are the invariable template patch and the current search frame. The static template information lacks the perception of dynamic feature information flow, and the shallow feature extraction and linear sequential mapping severely limit the mining of feature expressiveness. This makes it difficult for many existing SNs to cope with the challenges of UAV tracking, such as scale variation and viewpoint change caused by the change in height and angle of the UAV, and the challenges of background clutter and occlusion caused by complex aviation backgrounds. Secondly, the SNs trackers for UAV tracking still struggle with extracting lightweight and effective features. A tracker with a heavy-weighted backbone is not welcome due to the limited computing power of the UAV platform. Therefore, we propose a lightweight spatial-temporal contextual Siamese tracking system for UAV tracking (SiamST). The proposed SiamST improves the UAV tracking performance by augmenting the horizontal spatial information and introducing vertical temporal information to the Siamese network. Specifically, a high-order multiscale spatial module is designed to extract multiscale remote high-order spatial information, and a temporal template transformer introduces temporal contextual information for dynamic template updating. The evaluation and contrast results of the proposed SiamST with many state-of-the-art trackers on three UAV benchmarks show that the proposed SiamST is efficient and lightweight.

As an important branch of remote sensing detection, aerial tracking has been widely used in many fields. However, due to the unique aviation perspective, aerial tracking faces more challenges than general tracking. To achieve accurate and robust aerial tracking, many trackers choose transformer structure to improve the performance of the network. However, the high amount of computation of transformer does not meet the lightweight requirements of aerial tracking. Therefore, this letter designs a lightweight high-order transformer (HOT). The proposed HOT includes an encoder and a decoder. The encoder uses efficient recursive convolution to realize high-order interaction of information and improve the feature extraction ability of the network while greatly reducing the amount of computation. The decoder designs multiple cross-correlation structure to perform the deep fusion of the template branch and search branch. We integrate the proposed HOT into the Siamese-based network (SiamHOT) for aerial tracking. Finally, we evaluated the proposed SiamHOT in three aerial tracking benchmarks such as UAV20L, UAV123@10fps and DTB70, the results show that the proposed SiamHOT achieves superior performance compared to state-of-the-art (SOTA) trackers. Besides, we conducted the speed evaluation on RTX3060Ti, and the results show that the proposed SiamHOT achieves 116fps running speed, which is better than the comparison trackers.

Benefiting from the powerful feature extraction capability of deep learning, the Siamese tracker stands out due to its advanced tracking performance. However, constrained by the complex backgrounds of aerial tracking, such as low resolution, occlusion, similar objects, small objects, scale variation, aspect ratio change, deformation and limited computational resources, efficient and accurate aerial tracking is still difficult to realize. In this work, we design a lightweight and efficient adaptive temporal contextual aggregation Siamese network for aerial tracking, which is designed with a parallel atrous module (PAM) and adaptive temporal context aggregation model (ATCAM) to mitigate the above problems. Firstly, by using a series of atrous convolutions with different dilation rates in parallel, the PAM can simultaneously extract and aggregate multi-scale features with spatial contextual information at the same feature map, which effectively improves the ability to cope with changes in target appearance caused by challenges such as aspect ratio change, occlusion, scale variation, etc. Secondly, the ATCAM adaptively introduces temporal contextual information to the target frame through the encoder-decoder structure, which helps the tracker resist interference and recognize the target when it is difficult to extract high-resolution features such as low-resolution, similar objects. Finally, experiments on the UAV20L, UAV123@10fps and DTB70 benchmarks demonstrate the impressive performance of the proposed network running at a high speed of over 75.5 fps on the NVIDIA 3060Ti.

In this paper, we address aerial tracking tasks by designing multi-phase aware networks to obtain rich long-range dependencies. For aerial tracking tasks, the existing methods are prone to tracking drift in scenarios with high demand for multi-layer long-range feature dependencies such as viewpoint change caused by the characteristics of the UAV shooting perspective, low resolution, etc. In contrast to the previous works that only used multi-scale feature fusion to obtain contextual information, we designed a new architecture to adapt the characteristics of different levels of features in challenging scenarios to adaptively integrate regional features and the corresponding global dependencies information. Specifically, for the proposed tracker (SiamMAN), we first propose a two-stage aware neck (TAN), where first a cascaded splitting encoder (CSE) is used to obtain the distributed long-range relevance among the sub-branches by the splitting of feature channels, and then a multi-level contextual decoder (MCD) is used to achieve further global dependency fusion. Finally, we design the response map context encoder (RCE) utilizing long-range contextual information in backpropagation to accomplish pixel-level updating for the deeper features and better balance the semantic and spatial information. Several experiments on well-known tracking benchmarks illustrate that the proposed method outperforms SOTA trackers, which results from the effective utilization of the proposed multi-phase aware network for different levels of features.

Siamese-based trackers have been widely used in object tracking. However, aerial remote tracking suffers from various challenges such as scale variation, viewpoint change, background clutter and occlusion, while most existing Siamese trackers are limited to single-scale and local features, making it difficult to achieve accurate aerial tracking. We propose the global multi-scale optimization and prediction head attentional Siamese network to solve this problem and improve aerial tracking performance. Firstly, a transformer-based multi-scale and global feature encoder (TMGFE) is proposed to obtain global multi-scale optimization of features. Then, the prediction head attentional module (PHAM) is proposed to add context information to the prediction head by adaptively adjusting the spatial position and channel contribution of the response map. Benefiting from these two components, the proposed tracker solves these challenges of aerial remote sensing tracking to some extent and improves tracking performance. Additionally, we conduct ablation experiments on aerial tracking benchmarks, including UAV123, UAV20L, UAV123@10fps and DTB70, to verify the effectiveness of the proposed network. The comparisons of our tracker with several state-of-the-art (SOTA) trackers are also conducted on four benchmarks to verify its superior performance. It runs at 40.8 fps on the GPU RTX3060ti.

For the Siamese network-based trackers utilizing modern deep feature extraction networks without taking full advantage of the different levels of features, tracking drift is prone to occur in aerial scenarios, such as target occlusion, scale variation, and low-resolution target tracking. Additionally, the accuracy is low in challenging scenarios of visual tracking, which is due to the imperfect utilization of features. To improve the performance of the existing Siamese tracker in the above-mentioned challenging scenes, we propose a Siamese tracker based on Transformer multi-level feature enhancement with a hierarchical attention strategy. The saliency of the extracted features is enhanced by the process of Transformer Multi-level Enhancement; the application of the hierarchical attention strategy makes the tracker adaptively notice the target region information and improve the tracking performance in challenging aerial scenarios. Meanwhile, we conducted extensive experiments and qualitative or quantitative discussions on UVA123, UAV20L, and OTB100 datasets. Finally, the experimental results show that our SiamHAS performs favorably against several state-of-the-art trackers in these challenging scenarios.

Siamese network-based trackers satisfy the balance between performance and efficiency for visual tracking. However, they do not have enough robustness to handle the challenges of target occlusion and similar objects. In order to improve the robustness of the tracking algorithm, this paper proposes visual tracking with FPN based on Transformer and response map enhancement. In this paper, a feature pyramid structure based on Transformer is designed to encode robust target-specific appearance features, as well as the response map enhanced module to improve the tracker’s ability to distinguish object and background. Extensive experiments and ablation experiments are conducted on many challenging benchmarks such as UAV123, GOT-10K, LaSOT and OTB100. These results show that the tracking algorithm we proposed in this paper can effectively improve the tracking robustness against the challenges of target occlusion and similar object, and thus improve the precision rate and success rate of the tracking algorithm.