A Novel Anti-Drift Visual Object Tracking Algorithm Based on Sparse Response and Adaptive Spatial-Temporal Context-Aware

Abstract

Correlation filter (CF) based trackers have gained significant attention in the field of visual single-object tracking, owing to their favorable performance and high efficiency; however, existing trackers still suffer from model drift caused by boundary effects and filter degradation. In visual tracking, long-term occlusion and large appearance variations easily cause model degradation. To remedy these drawbacks, we propose a sparse adaptive spatial-temporal context-aware method that effectively avoids model drift. Specifically, a global context is explicitly incorporated into the correlation filter to mitigate boundary effects. Subsequently, an adaptive temporal regularization constraint is adopted in the filter training stage to avoid model degradation. Meanwhile, a sparse response constraint is introduced to reduce the risk of further model drift. Furthermore, we apply the alternating direction multiplier method (ADMM) to derive a closed-solution of the object function with a low computational cost. In addition, an updating scheme based on the APCE-pool and Peak-pool is proposed to reveal the tracking condition and ensure updates of the target’s appearance model with high-confidence. The Kalam filter is adopted to track the target when the appearance model is persistently unreliable and abnormality occurs. Finally, extensive experimental results on OTB-2013, OTB-2015 and VOT2018 datasets show that our proposed tracker performs favorably against several state-of-the-art trackers.

Global Multi-Scale Optimization and Prediction Head Attentional Siamese Network for Aerial Tracking

“…where ReLu means the activation function, 𝑊 and 𝑏 (i = 1, 2) represent para weight and biases. Taking 𝐿 as an example, we use 𝐿 as query, 𝐿 as the ke value, then we use formulation (5) to calculate the global information interrelati tween 𝐿 and 𝐿 . The result is added to 𝐿 as residual connection, then we apply normalization and feed-forward and layer-normalization again to establish inform interaction between 𝐿 and 𝐿 .…”

“…where ReLu means the activation function, W i and b i (I = 1, 2) represent parameter weight and biases. Taking L 1 as an example, we use L 2 as query, L 1 as the key and value, then we use formulation (5) to calculate the global information interrelation between L 2 and L 1 .…”

“…However, in aerial remote sensing tracking, there are many challenges, such as scale variation, viewpoint change, occlusion, similar object and background clutter. These challenges make accurate and robust aerial tracking still a challenging task [5].…”

Lightweight Spatial-Temporal Contextual Aggregation Siamese Network for Unmanned Aerial Vehicle Tracking

“…Object tracking is an important computer vision task in which a tracker receives the target information in the first frame of a video and computationally recognizes the target in the following video frames. As a branch of tracking task, aerial tracking is one of the important tasks in remote sensing detection, which is applied in many fields, such as intelligent surveillance [1,2], aerial detection [3], motion object analysis [4,5], etc. Therefore, many researches focus on aerial tracking [6][7][8][9][10].…”

SiamHAS: Siamese Tracker with Hierarchical Attention Strategy for Aerial Tracking

“…Correctly distinguishing the foreground from the background is the key factor to locate the target object location and generate the predicted banding box. A step further, according to the implementation principle, it can be divided into correlation filter-based [14][15][16][17] and deep learning-based [18][19][20][21] tracking models. The key step in the correlation filter approach is to obtain the filter with optimal performance so that the search area and the target area could generate an accurate maximum response.…”