Current Frame Research Articles

Experimental Study on Seismic Behavior of a New Separately‐Anchored Self‐Centering Beam‐Column Connection

ABSTRACTPost‐tensioned self‐centering (PTSC) structures that use PT tendons to assemble precast members and provide structural resistance have demonstrated superior seismic performance and self‐centering capacity. Current PTSC frames generally serially connect all beams in a floor. This assembly method would induce interferences among different spans and increase the risk of progressive failure once local damage occurs in a bay. This paper proposes a new type of PTSC joint to avoid such serial connection of beams. By employing steel beams to serve as the anchorages of PT tendons, each bay of the frame is separately prestressed. The composite prestressed beams can be readily fixed to columns by high‐strength bolts, hence facilitating the construction of such separately‐anchored self‐centering (SASC) frame building. External friction dampers are installed to enhance the energy dissipation capacity of SASC connections. Eight full‐scale cyclic loading tests are performed to compare the performance of the new SASC connections with conventional PTSC joints and also comprehensively evaluate the seismic behavior of SASC connections with different parameters. The test results validate the excellent damage mitigation abilities of both types of connections while the superior mechanical performance of the new SASC connections. In addition, the effects of key design parameters such as initial prestress level, damper force, and tendon number on the seismic behavior of the SASC connection are evaluated, and design recommendations for the application of the proposed precast beam‐column connection are also provided.

Spatial-Temporal Graph Enhanced DETR Towards Multi-Frame 3D Object Detection.

The Detection Transformer (DETR) has revolutionized the design of CNN-based object detection systems, showcasing impressive performance. However, its potential in the domain of multi-frame 3D object detection remains largely unexplored. In this paper, we present STEMD, a novel end-to-end framework that enhances the DETR-like paradigm for multi-frame 3D object detection by addressing three key aspects specifically tailored for this task. First, to model the inter-object spatial interaction and complex temporal dependencies, we introduce the spatial-temporal graph attention network, which represents queries as nodes in a graph and enables effective modeling of object interactions within a social context. To solve the problem of missing hard cases in the proposed output of the encoder in the current frame, we incorporate the output of the previous frame to initialize the query input of the decoder. Finally, it poses a challenge for the network to distinguish between the positive query and other highly similar queries that are not the best match. And similar queries are insufficiently suppressed and turn into redundant prediction boxes. To address this issue, our proposed IoU regularization term encourages similar queries to be distinct during the refinement. Through extensive experiments, we demonstrate the effectiveness of our approach in handling challenging scenarios, while incurring only a minor additional computational overhead.

Contour-based object forecasting for autonomous driving

A novel algorithm, called contour-based object forecasting (COF), to simultaneously perform contour-based segmentation and depth estimation of objects in future frames in autonomous driving systems is proposed in this paper. The proposed algorithm consists of encoding, future forecasting, decoding, and 3D rendering stages. First, we extract the features of observed frames, including past and current frames. Second, from these causal features, we predict the features of future frames using the future forecast module. Third, we decode the predicted features into contour and depth estimates. We obtain object depth maps aligned with segmentation masks via the depth completion using the predicted contours. Finally, from the prediction results, we render the forecasted objects in a 3D space. Experimental results demonstrate that the proposed algorithm reliably forecasts the contours and depths of objects in future frames and that the 3D rendering results intuitively visualize the future locations of the objects.

Joint Object Detection and Multi-Object Tracking Based on Hypergraph Matching

Addressing the challenges in online multi-object tracking algorithms under complex scenarios, where the independence among feature extraction, object detection, and data association modules leads to both error accumulation and the difficulty of maintaining visual consistency for occluded objects, we have proposed an end-to-end multi-object tracking method based on hypergraph matching (JDTHM). Initially, a feature extraction and object detection module is introduced to achieve preliminary localization and description of the objects. Subsequently, a deep feature aggregation module is designed to extract temporal information from historical tracklets, amalgamating features from object detection and feature extraction to enhance the consistency between the current frame features and the tracklet features, thus preventing identity swaps and tracklet breaks caused by object detection loss or distortion. Finally, a data association module based on hypergraph matching is constructed, integrating with object detection and feature extraction into a unified network, transforming the data association problem into a hypergraph matching problem between the tracklet hypergraph and the detection hypergraph, thereby achieving end-to-end model optimization. The experimental results demonstrate that this method has yielded favorable qualitative and quantitative analysis results on three multi-object tracking datasets, thereby validating its effectiveness in enhancing the robustness and accuracy of multi-object tracking tasks.

A proposal-level class-aware graph convolutional network and memory bank for thyroid nodule detection in ultrasound videos

Thyroid nodule detection in ultrasound is an effective method for early thyroid cancer diagnosis and can significantly reduce the workload of radiologists. However, this technology still faces considerable challenges due to issues such as low image quality, image artifacts, and speckle noise in ultrasound images. In this paper, we design a new framework for thyroid nodule detection, which enhances low-quality features in ultrasound videos for addressing thyroid nodule detection in ultrasound videos by leveraging high-quality features. In our framework, we propose a novel proposal-level class-aware graph convolutional network module, which removes noise interference from different classes and effectively utilizes temporal information from multiple frames to improve feature representation of the current frames. Furthermore, to further enhance the detection capability of the network for thyroid nodules in ultrasound videos, we design a new proposal-level memory bank to store and update high-quality proposal features in ultrasound videos. By fusing high-quality features from the memory bank with features of the current frames, our approach enables the enhancement of low-quality features in the current frames, thereby improving the performance of the network. Experimental results demonstrate that our proposed framework achieves a significant improvement over the previous state-of-the-art methods and superior real-time inference speed on our collected ultrasound thyroid video dataset.

Motion-Aware Self-Supervised RGBT Tracking with Multi-Modality Hierarchical Transformers

Supervised RGBT (SRGBT) tracking tasks need both expensive and time-consuming annotations. Therefore, the implementation of Self-Supervised RGBT (SSRGBT) tracking methods has become increasingly important. Straightforward SSRGBT tracking methods use pseudo-labels for tracking, but inaccurate pseudo-labels can lead to object drift, which severely affects tracking performance. This article proposes a self-supervised RGBT object tracking method (S2OTFormer) to bridge the gap between tracking methods supervised under pseudo-labels and ground truth labels. Firstly, to provide more robust appearance features for motion cues, we introduce a multi-modality hierarchical transformer (MHT) module for feature fusion. This module allocates weights to both modalities and strengthens the expressive capability of the MHT module through multiple nonlinear layers to fully utilize the complementary information of the two modalities. Secondly, in order to solve the problems of motion blur caused by camera motion and inaccurate appearance information caused by pseudo-labels, we introduce a motion-aware mechanism (MAM). The MAM extracts the average motion vectors from the previous multi-frame search frame features and constructs the consistency loss with the motion vectors of the current search frame features. The motion vectors of inter-frame objects are obtained by reusing the inter-frame attention map to predict coordinate positions. Finally, to further reduce the effect of inaccurate pseudo-labels, we propose an Attention-Based Multi-Scale Enhancement Module. By introducing cross-attention to achieve more precise and accurate object tracking, this module overcomes the receptive field limitations of traditional CNN tracking heads. We demonstrate the effectiveness of S2OTFormer on four large-scale public datasets through extensive comparisons as well as numerous ablation experiments. The source code is available at https://github.com/LiShenglana/S2OTFormer .

Intercropping grapevine with rosemary: A promising alternative to conventional agriculture in the current frame of global change

The Mediterranean region is well known for its long history of viticulture and the use of herbs and spices in its cuisine. Given the increasing focus on intercropping as a sustainable agricultural approach to maximize land use efficiency or improve soil fertility amid global change, we conducted a study to assess the feasibility of intercropping grapevine (Vitis vinifera L. cv. Merlot) with rosemary (Salvia rosmarinus Spenn.) in a Mediterranean vineyard. The trial was established in a 22-year-old vineyard, where rows of rosemary plants were cultivated between grapevine rows one year prior to the evaluation. To test intercropping feasibility and the abiotic stress effects on the harvest of rosemary, two sampling dates were set to monitor grape maturation (veraison and harvest) and rosemary quality in terms of antioxidant contents and composition, so that both crops were simultaneously analyzed under typical Mediterranean conditions, using both physiological and quality parameters. Results showed that intercropping with young rosemary plants did not affect grapevine physiology state, and observed differences in the two samplings were due to environmental stress only, indicating a positive coexistence between these crops in the context of global change. Regarding grape quality, variations in total soluble sugars, titratable acidity, and phenolic compounds were influenced by ripening during the summer, but not by intercropping, thus maintaining fruit quality. Rosemary contained high levels of antioxidants and exhibited strong antioxidant activity, particularly during mid-August (coinciding with vintage), indicating its potential as a valuable source of natural antioxidants for dietary intake, including phenolic diterpenes, vitamin C, and vitamin E. In conclusion, intercropping grapevines with rosemary did not negatively impact plant physiological performance or grape quality, promoting the production of two high-quality products with significant added value in the food sector under current stressful Mediterranean climatic conditions.

PZS‐Net: Incorporating of Frame Sequence and Multi‐Scale Priors for Prostate Zonal Segmentation in Transrectal Ultrasound

Transrectal ultrasound (TRUS) videos offer valuable histopathologic information about the prostate. Accurate prostate zonal segmentation in TRUS videos is vital for diagnosing prostate cancer and guiding surgery. However, TRUS videos are manually recorded by urologists, resulting in no standardized coordinate system, which limits direct prostate zonal segmentation in these videos. To overcome the limitation, a novel Prostate Zonal Segmentation Network (PZS‐Net), based on U‐Net, which learns critical cross‐frame information and multi‐scale features from sequential frames, is proposed. First, a sequential frame cross‐attention (SFCA) module is designed to capture remote information from sequential frames to enhance the feature representation of the current frame. The SFCA module is embedded at each skip connection layer to extract crucial cross‐frame information. Then, a multi‐scale fusion (MSF) module that utilizes three parallel branches with different atrous convolutions is designed. The MSF module is placed at the bottleneck layer to dynamically fuse multi‐scale context information from high‐level features. Extensive experiments on TRUS image datasets show that the PZS‐Net achieves higher accuracy in both the transitional zone (dice coefficient [Dice]: 68.90% ± 1.73%, mean intersection over union [mIoU]: 59.19% ± 2.09%, 95% Hausdorff distance [HD95]: 5.02 ± 0.83 mm) and the peripheral zone (Dice: 63.99% ± 3.16%, mIoU: 54.60% ± 3.35%, HD95: 5.28 ± 1.12 mm) and demonstrates the effectiveness and competitiveness of its key components via comprehensive ablation studies.

End-to-end multiple object tracking in high-resolution optical sensors of drones with transformer models

Drone aerial imaging has become increasingly important across numerous fields as drone optical sensor technology continues to advance. One critical challenge in this domain is achieving both accurate and efficient multi-object tracking. Traditional deep learning methods often separate object identification from tracking, leading to increased complexity and potential performance degradation. Conventional approaches rely heavily on manual feature engineering and intricate algorithms, which can further limit efficiency. To overcome these limitations, we propose a novel Transformer-based end-to-end multi-object tracking framework. This innovative method leverages self-attention mechanisms to capture complex inter-object relationships, seamlessly integrating object detection and tracking into a unified process. By utilizing end-to-end training, our approach simplifies the tracking pipeline, leading to significant performance improvements. A key innovation in our system is the introduction of a trajectory detection label matching technique. This technique assigns labels based on a comprehensive assessment of object appearance, spatial characteristics, and Gaussian features, ensuring more precise and logical label assignments. Additionally, we incorporate cross-frame self-attention mechanisms to extract long-term object properties, providing robust information for stable and consistent tracking. We further enhance tracking performance through a newly developed self-characteristics module, which extracts semantic features from trajectory information across both current and previous frames. This module ensures that the long-term interaction modules maintain semantic consistency, allowing for more accurate and continuous tracking over time. The refined data and stored trajectories are then used as input for subsequent frame processing, creating a feedback loop that sustains tracking accuracy. Extensive experiments conducted on the VisDrone and UAVDT datasets demonstrate the superior performance of our approach in drone-based multi-object tracking.

Anti-noise weld seam tracking algorithm based on spatio-temporal memory mechanism

In laser vision-based robotic welding, existing seam tracking methods are seriously affected by noise and lack effective utilization of spatio-temporal information during welding process. To address this, this paper proposes an anti-noise weld seam tracking algorithm based on spatio-temporal memory mechanism. First, a memory mechanism with welding history memory information is designed, which can better adapt to seam appearance changes and noise interference during welding. Subsequently, a laser stripe tracker based on spatio-temporal memory is constructed, it enables the current frame to adaptively search historical valid information to eliminate effects of noise and appearance changes. The experimental results show that the proposed method has an average response speed of 43 FPS and an average tracking error of 1.35 pixels for various weld seams. The actual tracking error is less than 0.6 mm. The introduction of memory mechanism improves the weld tracking accuracy by 52.1 % compared to a single tracking template.

Enhanced neural video compression for cloud gaming videos with aligned frame generation

The burgeoning popularity of cloud gaming makes it critical for efficient video compression to relieve the growing bandwidth pressure. While existing neural video coding approaches have demonstrated strong compression potential on natural videos, there is an absence of efficient neural codecs dedicated to gaming videos. To bridge this gap, in this paper, we propose an end-to-end neural video compression method designed specifically for cloud gaming videos. By effectively utilizing the unique camera motion information inherent to cloud gaming, the previous reconstructed frame is maximally aligned to the current frame through a learning-based module with multiple losses, which then replaces the previous reconstructed frame for optical flow estimation. By significantly reducing the displacement between two consecutive frames caused by camera motion, the motion estimation accuracy is enhanced, effectively handling the large and abrupt motion scenarios frequently present in gaming videos. Furthermore, the aligned tensor obtained in the previous step is used to enhance the latent prior of the entropy model, providing a superior temporal prior for coding. Extensive experimental results demonstrate the superior performance of our proposed method compared to one of the previous state-of-the-art approaches, DCVC-HEM, providing significant progress in end-to-end neural compression in cloud gaming videos.

Temporal relation transformer for robust visual tracking with dual-memory learning

Recently, transformer trackers mostly associate multiple reference images with the search area to adapt to the changing appearance of the target. However, they ignore the learned cross-relations between the target and surrounding, leading to difficulties in building coherent contextual models for specific target instances. This paper presents a Temporal Relation Transformer Tracker (TRTT) for robust visual tracking, providing a concise approach to modeling temporal relations by dual target memory learning. Specifically, a temporal relation transformer network generates paired memories based on static and dynamic templates, which are reinforced interactively. The memory contains implicit relation hints that capture the relations between the tracked object and its immediate surroundings. More importantly, to ensure consistency of target instance identities between frames, the relation hints from previous frames are transferred to the current frame for merging temporal contextual attention. Our method also incorporates mechanisms for reusing favorable cross-relations and instance-specific features, thereby overcoming background interference in complex spatio-temporal interactions through a sequential constraint. Furthermore, we design a memory token sparsification method that leverages the key points of the target to eliminate interferences and optimize attention calculations. Extensive experiments demonstrate that our method surpasses advanced trackers on 8 challenging benchmarks while maintaining real-time running speed.

A new approach to video summary generation

This paper describes a video summary approach for an actual video file that provides a rapid way to assess its information. This is a difficult task to identify relevant or informative sections. When the original video demands knowledge of its content, we build a video summary via extracting in-depth characteristics of each original video. Video summarization deals with the generation of a condensed version of the original video by including meaningful frames or segments while eliminating redundant information provides a concise and accurate representation of the original video clips by providing the most representative summary with increased interest through the application of a matching function between the current and prior frames. The primary goal of this work is to give a strategy for recording video clips using security cameras and converting them into a short video to provide a clear analysis of the video by reducing excess and extracting the key video contents into the real video. In a lengthy video clip, viewers may not have enough time to see the full thing. The viewer may be interested in seeing a specific topic under the label "Important". It may be composed of pieces that may be shared on social media and communicate the semantics of long videos, particularly video snippets that must be submitted to Internet. To browse, will be needed a network with adequate bandwidth. That is, non-affiliated video streaming, indexing, and retrieval of video content makes it valuable to viewers.

An Improved Laplace Satellite Tracking Method Based on the Kalman Filter

When photoelectric measuring equipment is used to track satellites, the extraction of the short-term or long-term target often fails because the target is weak, clouds block the target, and/or the sun’s angle is too small, resulting in the loss of the tracking target. In this study, an improved Laplacian satellite tracking method based on the Kalman filter is proposed. Firstly, the improved Laplacian algorithm was used for the initial fitting of the equation of motion of a small amount of measurement data. Judgment of the validity and Kalman filtering was carried out on the current frame’s measurement data to calculate the optimal estimate of the current frame’s orbit data, and the accurate equation of motion was iteratively fitted to obtain high-precision data for predicting the satellite’s orbit frame by frame. Numerical tracking of the equipment was carried out. This method was experimentally validated on an actual optical measurement device. The test results showed that this method can make up for the frequent loss of short-term targets. Under the condition that the maximum deviation is less than 3”, the length of extrapolated data can be up to 30 s and the length of the measurement data was less than 30 s. This method may improve the stability of tracking equipment as well as the accuracy and integrity of the measurement data.

基于扫描上下文和 NDT-ICP 融合方案的果园同步定位与绘图方法研究

Simultaneous localization and mapping (SLAM) is one of the key technologies for agricultural robots to build maps and localize in complex orchard environments and realize unmanned autonomous operations. Due to the complexity of the orchard environment, the single canopy feature and the diffuse reflection of light caused by the leaves, etc., the map construction process of the orchard environment leads to mismatch and increases the cumulative error of the map construction. Aiming at the above problems, this paper propose a navigation map construction method for orchard environment based on the fusion of Scan Context and NDT-ICP. The method firstly searches the Ring key quickly to get the candidate frames, and scores the similarity between the candidate frames and the current frame, and effectively detects the loopbacks by two-stage searching algorithm to reduce the false matches in the map of orchard environment. Meanwhile, a point cloud alignment method based on the fusion of normal distribution transform coarse alignment and iterative nearest point exact alignment is used to reduce the cumulative error of the orchard environment map. The results show that the improved algorithm compensates the drift of the point cloud map with higher mapping accuracy, better real-time performance, lower resource utilization, higher overlap between the trajectory estimation and the real trajectory, smoother loops, and a 4% reduction in CPU occupancy. In the complex orchard environment, the root mean square error and standard deviation of the trajectories of this paper's algorithm are 0.57 m and 0.19 m, which are 68% and 83% higher than those of the loop detection algorithms in the Lightweight Ground Optimized Lidar Trajectory Measurement and Multivariate Terrain Mapping (LeGO-LOAM), respectively. Accurate map construction and low drift pose estimation can be performed.The research algorithm effectively reduces the influence of mis-matching and large cumulative error in the process of map construction in the orchard environment, meets the demand for high-precision environmental mapping in the orchard environment, and provides technical support for promoting unmanned operation in the orchard environment.

Tropical cyclone tracking from geostationary infrared satellite images using deep learning techniques

ABSTRACT Accurate tracking of tropical cyclones (TCs) can provide regions of interest for intelligent forecasting of TC tracks and intensity. There have been few studies on algorithms for automatic TC tracking. This study proposes an effective TC tracking method based on deep learning combined with infrared satellite images. The study first constructed a TC tracking dataset based on the infrared images of the China Fengyun-2D geostationary satellite covering six different TC intensity levels between 2009 and 2012. This included 47 complete cases (video sequences) of TCs from generation to extinction. Based on deep learning, the visual tracking algorithm SiamRPN was used as the model framework. Combining Bi-GRU and TC cloud spatiotemporal evolution characteristics to improve the performance of the SiamRPN network, the SiamTCNet target-tracking model was designed to track TCs automatically. Considering that the shape and scale of TC changes with time, a TC is regarded as a typical non-rigid object with obvious timing characteristics, so the first frame of a TC video sequence is combined with the satellite images of the first three frames of the current frame as inputs to the proposed SiamTCNet model, which then extracts the evolution of the TC’s spatial structure and its bidirectional temporal change information. The experimental results show that the TC tracking of the proposed model is a significant improvement over the original SiamRPN model.

Spatial-Temporal Contextual Aggregation Siamese Network for UAV Tracking

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.

Dual temporal memory network with high-order spatio-temporal graph learning for video object segmentation

Typically, Video Object Segmentation (VOS) always has the semi-supervised setting in the testing phase. The VOS aims to track and segment one or several target objects in the following frames in the sequence, only given the ground-truth segmentation mask in the initial frame. A fundamental issue in VOS is how to best utilize the temporal information to improve the accuracy. To address the aforementioned issue, we provide an end-to-end framework that simultaneously extracts long-term and short-term historical sequential information to current frame as temporal memories. The integrated temporal architecture consists of a short-term and a long-term memory modules. Specifically, the short-term memory module leverages a high-order graph-based learning framework to simulate the fine-grained spatial–temporal interactions between local regions across neighboring frames in a video, thereby maintaining the spatio-temporal visual consistency on local regions. Meanwhile, to relieve the occlusion and drift issues, the long-term memory module employs a Simplified Gated Recurrent Unit (S-GRU) to model the long evolutions in a video. Furthermore, we design a novel direction-aware attention module to complementarily augment the object representation for more robust segmentation. Our experiments on three mainstream VOS benchmarks, containing DAVIS 2017, DAVIS 2016, and Youtube-VOS, demonstrate that our proposed solution provides a fair tradeoff performance between both speed and accuracy.

TS-BEV: BEV object detection algorithm based on temporal-spatial feature fusion

In order to accurately identify occluding targets and infer the motion state of objects, we propose a Bird’s-Eye View Object Detection Network based on Temporal-Spatial feature fusion (TS-BEV), which replaces the previous multi-frame sampling method by using the cyclic propagation mode of historical frame instance information. We design a new Temporal-Spatial feature fusion attention module, which fully integrates temporal information and spatial features, and improves the inference and training speed. In response to realize multi-frame feature fusion across multiple scales and views, we propose an efficient Temporal-Spatial deformable aggregation module, which performs feature sampling and weighted summation from multiple feature maps of historical frames and current frames, and makes full use of the parallel computing capabilities of GPUs and AI chips to further improve efficiency. Furthermore, in order to solve the lack of global inference in the context of temporal-spatial fusion BEV features and the inability of instance features distributed in different locations to fully interact, we further design the BEV self-attention mechanism module to perform global operation of features, enhance global inference ability and fully interact with instance features. We have carried out extensive experimental experiments on the challenging BEV object detection nuScenes dataset, quantitative results show that our method achieves excellent performance of 61.5% mAP and 68.5% NDS in camera-only 3D object detection tasks, and qualitative results show that TS-BEV can effectively solve the problem of 3D object detection in complex traffic background with lack of light at night, with good robustness and scalability.

Spatial attention inference model for cascaded siamese tracking with dynamic residual update strategy

Target representation is crucial for visual tracking. Most Siamese-based trackers try their best to establish target models by using various deep networks. However, they neglect the exploration of correlation among features, which leads to the inability to learn more representative features. In this paper, we propose a spatial attention inference model for cascaded Siamese tracking with dynamic residual update strategy. First, a spatial attention inference model is constructed. The model fuses interlayer multi-scale features generated by dilation convolution to enhance the spatial representation ability of features. On this basis, we use self-attention to capture interaction between target and context, and use cross-attention to aggregate interdependencies between target and background. The model infers potential feature information by exploiting the correlations among features for building better appearance models. Second, a cascaded localization-aware network is introduced to bridge a gap between classification and regression. We propose an alignment-aware branch to resample and learn object-aware features from the predicted bounding boxes for obtaining localization confidence, which is used to correct the classification confidence by weighted integration. This cascaded strategy alleviates the misalignment problem between classification and regression. Finally, a dynamic residual update strategy is proposed. This strategy utilizes the Context Fusion Network (CFNet) to fuse the templates of historical and current frames to generate the optimal templates. Meanwhile, we use a dynamic threshold function to determine when to update by judging the tracking results. The strategy uses temporal context to fully explore the intrinsic properties of the target, which enhances the adaptability to changes in the target’s appearance. We conducted extensive experiments on seven tracking benchmarks, including OTB100, UAV123, TC128, VOT2016, VOT2018, GOT10k and LaSOT, to validate the effectiveness of our proposed algorithm.