Target Scale Changes Research Articles

EA‐YOLO: An Efficient and Accurate UAV Image Object Detection Algorithm

An improved EA‐YOLO object detection algorithm based on YOLOv5 is proposed to address the issues of drastic changes in target scale, low detection accuracy, and high miss rate in unmanned aerial vehicle aerial photography scenarios. Firstly, a DFE module was proposed to improve the effectiveness of feature extraction and enhance the whole model's ability to learn residual features. Secondly, a CWFF architecture was introduced to enable deeper feature fusion and improve the effectiveness of feature fusion. Finally, in order to solve the traditional algorithm's shortcomings it is difficult to detect small targets. We have designed a novel SDS structure and adopted a strategy of reusing low‐level feature maps to enhance the network's ability to detect small targets, making it more suitable for detecting some small objects in drone images. Experiments in the VisDrone2019 dataset demonstrated that the proposed EA‐YOLOs achieved an average accuracy mAP@0.5 of 39.9%, which is an 8% improvement over YOLOv5s, and mAP@0.5:0.95 of 22.2%, which is 5.2% improvement over the original algorithm. Compared with YOLOv3, YOLOv5l, and YOLOv8s, the mAP@0.5 of EA‐YOLOs improved by 0.9%, 1.8%, and 0.6%, while the GFLOPs decreased by 86.4%, 80.6%, and 26.7%. © 2024 Institute of Electrical Engineers of Japan and Wiley Periodicals LLC.

A Robust AR-DSNet Tracking Registration Method in Complex Scenarios

A robust AR-DSNet (Augmented Reality method based on DSST and SiamFC networks) tracking registration method in complex scenarios is proposed to improve the ability of AR (Augmented Reality) tracking registration to distinguish target foreground and semantic interference background, and to address the issue of registration failure caused by similar target drift when obtaining scale information based on predicted target positions. Firstly, the pre-trained network in SiamFC (Siamese Fully-Convolutional) is utilized to obtain the response map of a larger search area and set a threshold to filter out the initial possible positions of the target; Then, combining the advantage of the DSST (Discriminative Scale Space Tracking) filter tracker to update the template online, a new scale filter is trained after collecting multi-scale images at the initial possible position of target to reason the target scale change. And linear interpolation is used to update the correlation coefficient to determine the final position of target tracking based on the difference between two frames. Finally, ORB (Oriented FAST and Rotated BRIEF) feature detection and matching are performed on the accurate target position image, and the registration matrix is calculated through matching relationships to overlay the virtual model onto the real scene, achieving enhancement of the real world. Simulation experiments show that in complex scenarios such as similar interference, target occlusion, and local deformation, the proposed AR-DSNet method can complete the registration of the target in AR 3D tracking, ensuring real-time performance while improving the robustness of the AR tracking registration algorithm.

Explicit Visual Prompts for Visual Object Tracking

How to effectively exploit spatio-temporal information is crucial to capture target appearance changes in visual tracking. However, most deep learning-based trackers mainly focus on designing a complicated appearance model or template updating strategy, while lacking the exploitation of context between consecutive frames and thus entailing the when-and-how-to-update dilemma. To address these issues, we propose a novel explicit visual prompts framework for visual tracking, dubbed EVPTrack. Specifically, we utilize spatio-temporal tokens to propagate information between consecutive frames without focusing on updating templates. As a result, we cannot only alleviate the challenge of when-to-update, but also avoid the hyper-parameters associated with updating strategies. Then, we utilize the spatio-temporal tokens to generate explicit visual prompts that facilitate inference in the current frame. The prompts are fed into a transformer encoder together with the image tokens without additional processing. Consequently, the efficiency of our model is improved by avoiding how-to-update. In addition, we consider multi-scale information as explicit visual prompts, providing multiscale template features to enhance the EVPTrack's ability to handle target scale changes. Extensive experimental results on six benchmarks (i.e., LaSOT, LaSOText, GOT-10k, UAV123, TrackingNet, and TNL2K.) validate that our EVPTrack can achieve competitive performance at a real-time speed by effectively exploiting both spatio-temporal and multi-scale information. Code and models are available at https://github.com/GXNU-ZhongLab/EVPTrack.

Object Detection of UAV Aerial Image based on YOLOv8

With the development of technology, unmanned aerial vehicles (UAVs) have shed their military uses and gradually expanded to civilian and commercial fields. With the development of drone technology, object detection technology based on deep learning has become an important research topic in the field of drone applications. Apply object detection technology to unmanned aerial vehicles to achieve object detection and recognition of ground scenes from an aerial perspective. However, in aerial images taken by drones, the detection objects are mostly small targets, and the target scale changes greatly due to the influence of aerial perspective; The image background is complex, and the target object is easily occluded. It has brought many challenges to the target detection of unmanned aerial vehicles. Conventional object detection algorithms cannot guarantee detection accuracy when applied to drones, and optimizing the target detection performance of drones has become an important research topic in the field of drone applications. We improve the WIoUv3 loss function on the basis of YOLOv8s to reduce regression localization loss during training and improve the regression accuracy of the model. The experimental results indicate that the improved model mAP@0.5 It increased by 0.6 percentage points to 40.7%.

YOLOv5-OCDS: An Improved Garbage Detection Model Based on YOLOv5

As the global population grows and urbanization accelerates, the garbage that is generated continues to increase. This waste causes serious pollution to the ecological environment, affecting the stability of the global environmental balance. Garbage detection technology can quickly and accurately identify, classify, and locate many kinds of garbage to realize the automatic disposal and efficient recycling of waste, and it can also promote the development of a circular economy. However, the existing garbage detection technology has some problems, such as low precision and a poor detection effect in complex environments. Although YOLOv5 has achieved good results in garbage detection, the detection results cannot meet the requirements in complex scenarios, so this paper proposes a garbage detection model, YOLOv5-OCDS, based on an improved YOLOv5. Replacing the partial convolution in the neck with Omni-Dimensional Dynamic Convolution (ODConv) improves the expressiveness of the model. The C3DCN structure is constructed, and parts of the C3 structures in the neck are replaced by C3DCN structures, allowing the model to better adapt to object deformation and target scale change. The decoupled head is used for classification and regression tasks so that the model can learn each class’s characteristics and positioning information more intently, and flexibility and extensibility can be improved. The Soft Non-Maximum Suppression (Soft NMS) algorithm can better retain the target’s information and effectively avoid the problem of repeated detection. The self-built garbage classification dataset is used for related experiments, and the mAP@50 of the YOLOv5-OCDS model is 5.3% higher than that of the YOLOv5s; the value of mAP@50:95 increases by 12.3%. In the experimental environment of this study, the model’s Frames Per Second (FPS) was 61.7 f/s. In practical applications, when we use some old GPU, such as the GTX1060, it can still reach 50.3 f/s, so that real-time detection can be achieved. Thus, the improved model suits garbage detection tasks in complex environments.

Target detection based on multi-scale feature fusion and cross-channel interactive attention mechanism

Aiming at the problems of complex background, target scale change and small target in aerial image detection, we propose a YOLOv5 target detection algorithm based on multi-scale feature fusion and cross-channel interactive attention mechanism. Including: M-PPM (Multi-scale pyramid pooling module) is designed as a replacement for the SPP (Spatial Pyramid Pooling) structure in YOLOv5, so as to make full use of different scale features to fuse global feature information; CCA (Cross-channel interactive attention mechanism) is designed to realize cross-channel information interaction and utilization, and enhance the network’s capability to generalize and fusion efficiency of small target features. Bi-directional Feature Pyramid Network (BiFPN) is utilized to solve scale difference problem in multi-target detection. The proposed algorithm’s experimental results is 2.3 % and 1.8 % higher than YOLOv5 on the VisDrone and UAVDT aerial data sets, respectively.

An end-to-end multiple side-outputs fusion deep supervision network based remote sensing image change detection algorithm

Change detection is an effective means of monitoring surface changes. In order to solve the problem of reducing detection accuracy in remote sensing image change detection methods that only focus on extracting deep semantic information while ignoring high-resolution shallow information, this paper proposes an end-to-end multi-side output fusion deep supervised network for high-resolution biphasic remote sensing image change detection. The enhanced feature extraction module is used to increase the Receptive field and capture multi-scale features. The dense skip connection module utilizes skip connections to fuse shallow position information and deep semantic information, alleviating the problem of feature information loss. The deep supervision module with multiple side-outputs fusion utilizes the information complementarity between different scale feature maps, combined with multi-scale prediction maps, to improve the robustness of target scale changes. The combined Loss function is used to alleviate the problem of imbalance between positive and negative samples, so that more attention is paid to the learning of changing characteristics in the process of training the network. The proposed method was validated on CDD dataset and DSIFN dataset, with F1 scores of 91.58% and 86.69%, respectively, which were improved by 1.00% and 5.53% compared to the suboptimal network.

Deformable Density Estimation via Adaptive Representation.

Crowd counting is the basic task of crowd analysis and it is of great significance in the field of public safety. Therefore, it receives more and more attention recently. The common idea is to combine the crowd counting task with convolutional neural networks to predict the corresponding density map, which is generated by filtering the dot labels with specific Gaussian kernels. Although the counting performance is promoted by the newly proposed networks, they all suffer one conjunct problem, which is due to the perspective effect, there is significant scale contrast among targets in different positions within one scene, but the existing density maps can not represent this scale change well. To address the prediction difficulties caused by target scale variation, we propose a scale-sensitive crowd density map estimation framework, which focuses on dealing with target scale change from density map generation, network design, and model training stage. It consists of the Adaptive Density Map (ADM), Deformable Density Map Decoder (DDMD), and Auxiliary Branch. To be specific, the Gaussian kernel size variates adaptively based on target size to generate ADM that contains scale information for each specific target. DDMD introduces the deformable convolution to fit the Gaussian kernel variation and boosts the model's scale sensitivity. The Auxiliary Branch guides the learning of deformable convolution offsets during the training phase. Finally, we construct experiments on different large-scale datasets. The results show the effectiveness of the proposed ADM and DDMD. Furthermore, the visualization demonstrates that deformable convolution learns the target scale variation.

A Fast Adaptive Multi-Scale Kernel Correlation Filter Tracker for Rigid Object.

The efficient and accurate tracking of a target in complex scenes has always been one of the challenges to tackle. At present, the most effective tracking algorithms are basically neural network models based on deep learning. Although such algorithms have high tracking accuracy, the huge number of parameters and computations in the network models makes it difficult for such algorithms to meet the real-time requirements under limited hardware conditions, such as embedded platforms with small size, low power consumption and limited computing power. Tracking algorithms based on a kernel correlation filter are well-known and widely applied because of their high performance and speed, but when the target is in a complex background, it still can not adapt to the target scale change and occlusion, which will lead to template drift. In this paper, a fast multi-scale kernel correlation filter tracker based on adaptive template updating is proposed for common rigid targets. We introduce a simple scale pyramid on the basis of Kernel Correlation Filtering (KCF), which can adapt to the change in target size while ensuring the speed of operation. We propose an adaptive template updater based on the Mean of Cumulative Maximum Response Values (MCMRV) to alleviate the problem of template drift effectively when occlusion occurs. Extensive experiments have demonstrated the effectiveness of our method on various datasets and significantly outperformed other state-of-the-art methods based on a kernel correlation filter.

An improved SSD method for infrared target detection based on convolutional neural network

Target detection is the basis for automatic target recognition system of infrared imaging guidance to complete subsequent tasks such as recognition and tracking. Existing systems have not the autonomous learning ability of target feature, and it will be powerless once the task environment exceeds the pre-planned condition. The single-stage target detection based on deep learning has the ability of autonomous learning and high computational efficiency, which is an effective way to solve the problem of infrared imaging guidance target detection in complex environment. SSD (Single Shot MultiBox Detector) is a classical single-stage detection model, however, the convolution layer with strong semantic information in SSD has low resolution, which is not conducive to small target detection. In addition, the location loss of SSD does not consider the impact of target scale change. Therefore, this paper puts forward two improvement ideas in view of SSD: (1) Starting from FPN (Feature Pyramid Network), feature channel’s importance is distinguished through efficient channel attention mechanism, the contribution of each feature layer to the fusion output is described based on the learnable weight, and the feature weighted fusion of bidirectional multi-scale is realized between the feature layer which has low resolution and strong semantics and the feature layer which has high resolution and weak semantics. (2) Starting from IoU (Intersection over Union) and considering non overlapping parts and geometric relationship between the predicted box and the ground-truth box, the location loss of SSD that remains invariable to the target scale change is constructed to improve the sensitivity of the detection model to the locating error of small target. The experimental results show that, for 300 × 300 input, the presented method achieves 84.7% mAP (mean Average Precision) on VOC2007 test and for 512 × 512 input, it reaches 86.6%. On the self-built infrared aircraft data set, the proposed method achieves 81.1% mAP and can detect more small targets. Without affecting detection speed, the presented method on experimental results outperforms some comparable state-of-the-art models such as YOLOv3 (You Only Look Once), DSSD (Deconvolutional Single Shot Multibox Detector), RSSD (Rainbow Single Shot Multibox Detector) and FSSD (Fusion Single Shot Multibox Detector).

SCFFNet: Spatial Context Feature Fusion Network for Understanding the Highly Congested Scenes

Accurate counting in dense scenes can effectively prevent the occurrence of abnormal events, which is crucial for flow management, traffic control, and urban safety. In recent years, the application of deep learning technology in counting tasks has significantly improved the performance of models, but it still faces many challenges, including the diversity of target distribution between image and background, the drastic change of target scale, and serious occlusion. To solve these problems, this paper proposes a spatial context feature fusion network, abbreviated as SCFFNet, to understand highly congested scenes and perform accurate counts as well as produce high-quality estimated density maps. SCFFNet first uses rich convolutions with different scales to calculate scale-aware features, adaptively encodes the scale of contextual information needed to accurately estimate density maps, and then calibrates and refuses the fused feature maps through a channel spatial attention-aware module, which improves the model’s ability to suppress background and focus on main features. Finally, the final estimated density map is generated by a dilated convolution module. We conduct experiments on five public crowd datasets, UCF_CC_50, WorldExpo’10, ShanghaiTech, Mall, and Beijing BRT, and the results show that our method achieves lower counting errors than existing state-of-the-art methods. In addition, we extend SCFFNet to count other objects, such as vehicles in the vehicle dataset HBR_YD, and the experimental results show that our proposed method significantly improves the output quality with higher accuracy than previous methods.

Multiscale anti-deformation network for target tracking in UAV aerial videos

With the popularity of unmanned aerial vehicle (UAV) technology, remote sensing target tracking in aerial videos from UAVs has drawn much attention. However, various problems occur, such as obvious target scale changes and frequent target shape updates in UAV aerial videos, and typical tracking algorithms have difficulties in solving these challenges. Therefore, we propose a remote sensing target tracking method for UAV videos based on a multiscale antideformation network (MSADN). This method uses the fully convolutional Siamese network (SiameseFC) as its basic architecture. First, in the target feature extraction stage, we use a receptive field block module to change the single convolutional layer of the network. And the multibranch structure can improve the algorithm’s adaptability to target scale changes. Then, in the tracking stage, to solve the problem of the lack of template updates, we integrate a template dynamic update module into the SiameseFC architecture. This module uses long short-term memory to generate control signals to dynamically update target shape information, and it will improve the algorithm’s ability to cope with frequent target shape updates. Finally, compared to state-of-the-art trackers, experimental results show that the MSADN algorithm can obtain better performance (75.9% Prec, 56.4% Succ) while ensuring higher efficiency (60 fps).

Parallel CNN Network Learning-Based Video Object Recognition for UAV Ground Detection

Video object recognition for UAV ground detection is widely used in target search, daily patrol, environmental reconnaissance, and other fields. So, we propose the novel parallel deep learning network with the ability of the global and local joint feature extraction for the UAV video target detection. This paper focuses on solving the problems of feature extraction and target background discrimination required by target discovery to realize target discovery. Break through the key problems of real-time target recognition, such as multiscale targets, high background complexity, many small targets, dense target arrangement, and multidirection, and put forward an optimized network scheme, aiming at the problem of multiscale of image target and aiming at the problem of large change of target scale in image. In the network, the corresponding targets with different sizes and different aspect ratios are matched to make the different targets match the closest, and then, the position of the detection box is fine-tuned by regression. For the special problem of image viewing angle and for the rotation invariance of the airborne down looking image of the target, the usual solution is through data enhancement; that is, through the rotation transformation of the training data, the neural network can learn the rotation invariance of the target. Aiming at the problem of multi-directional image target and aiming at the problems of large target aspect ratio, large target tilt angle, and changeable direction in the target, we propose to use the tilt detection frame instead of the ordinary rectangular detection frame. Aiming at the problem of dense arrangement of image targets and aiming at a large number of densely arranged targets in the image, a feature refining module is proposed, which can effectively improve the detection performance of the detector for densely arranged targets. The experimental results shows that the proposed algorithm achieves more than 10% on the target detection accuracy with focal length change of 1-10 times. The detection accuracy meets the requirements of practical application.

Network Fault Diagnosis of Embedded System Based on Topology Constraint and Data Mining.

Maintaining the safe and efficient operation of network technology is an important development task of the computer industry. Topology constraint can optimize and combine the tracking results and select the target objects with better tracking performance to obtain the final tracking results and determine the target scale changes. Data mining technology can reduce the number of combinations to be detected, reduce the workload, and improve the timeliness and accuracy of the process of mining alarm association rules. Therefore, based on the summary and analysis of previous research results, this paper studied the network fault diagnosis of the embedded system method based on topology constraint and data mining. Firstly, a fault diagnosis topology model was established by constructing a topology search algorithm, which eliminated the filtering of association rules without topology relationship; the association rule-based data mining model was analyzed through the collection of network alarm data; the model algorithm was applied to the simulation experiment of network fault diagnosis of the embedded system and achieved good results. The results show that correcting rage of retrieval varies from 0.65 to 090 under different window sizes; the running time of the proposed method drops from 310 s to 35 s during 1–8 step/s of the sliding step, while the node degree ranges from 8 to 14 and diagnostic accuracy ranges from 0.97 to 0.94; the remaining alarm number increases from 0.5 to 3.5 threshold value, while the regular association number distributed in an interval of 40 to 140. The algorithm in this paper provides a reference for further research on network fault diagnosis of the embedded system.

RPformer: A Robust Parallel Transformer for Visual Tracking in Complex Scenes

The Siamese architecture has shown remarkable performance in the field of visual tracking. Although the existing Siamese-based tracking methods have achieved a relative balance between accuracy and speed, the performance of many trackers in complex scenes is often unsatisfactory, which is mainly caused by interference factors, such as target scale changes, occlusion, and fast movement. In these cases, excessive trackers cannot employ sufficiently the target feature information and face the dilemma of information loss. In this work, we propose a novel parallel Transformer network architecture to achieve robust visual tracking. The proposed method designs the Transformer-1 module, the Transformer-2 module, and the feature fusion head (FFH) based on the attention mechanism. The Transformer-1 module and the Transformer-2 module are regarded as corresponding complementary branches in the parallel architecture. The FFH is used to integrate the feature information of the two parallel branches, which can efficiently exploit the feature dependence relationship between the template and the search region, and comprehensively explore rich contextual information. Finally, by combining the core ideas of Siamese and Transformer, we present a simple and robust tracking framework called RPformer, which does not require any prior knowledge and avoids the trouble of adjusting hyperparameters. Numerous experiments show that the proposed tracking method achieves more outstanding performance than the state-of-the-art trackers on seven tracking benchmarks, which can meet the real-time requirements at a running speed exceeding 50.0 frames/s.

Research on Target Tracking Algorithm of Twin Networks Integrating Attention Mechanism

Aiming at the current stage of the twin network target tracking algorithm, the tracking target is occluded, the tracking is affected by illumination, and the target's scale change from far to near or from near to far causes tracking failure. This article will optimize and improve from two directions. The twin neural network first uses an adaptive detailed feature extraction, adds a residual network to the twin network, and embeds a detailed feature retention module in each layer, amplifies the changes in the target feature, and retains the important structure of the original target feature Details: Secondly, the introduction of a spatial attention mechanism allows the main branch to pay more attention to the area to be matched, improves the ability to distinguish features, and makes the tracking effect better. In order to verify the effectiveness of this experiment, this experiment was tested on the data set OTB2015. The experiment proved that the proposed algorithm performs better in accuracy and success rate.

Vision-based target point tracking and aiming method

PurposeLaser absolute distance measurement has the characteristics of high precision, wide range and non-contact. In laser ranging system, tracking and aiming measurement point is the precondition of automatic measurement. To solve this problem, this paper aims to propose a novel method.Design/methodology/approachFor the central point of the hollow angle coupled mirror, this paper proposes a method based on correlation filtering and ellipse fitting. For non-cooperative target points, this paper proposes an extraction method based on correlation filtering and feature matching. Finally, a visual tracking and aiming system was constructed by combining the two-axis turntable, and experiments were carried out.FindingsThe target tracking algorithm has an accuracy of 91.15% and a speed of 19.5 frames per second. The algorithm can adapt to the change of target scale and short-term occlusion. The mean error and standard deviation of the center point extraction of the hollow Angle coupling mirror are 0.20 and 0.09 mm. The mean error and standard deviation of feature points matching for non-cooperative target were 0.06 mm and 0.16 mm. The visual tracking and aiming system can track a target running at a speed of 0.7 m/s, aiming error mean is 1.74 pixels and standard deviation is 0.67 pixel.Originality/valueThe results show that this method can achieve fast and high precision target tracking and aiming and has great application value in laser ranging.

Target Tracking Algorithm Based on Adaptive Scale Detection Learning

In this paper, to better solve the problem of low tracking accuracy caused by the sudden change of target scale, we design and propose an adaptive scale mutation tracking algorithm using a deep learning network to detect the target first and then track it using the kernel correlation filtering method and verify the effectiveness of the model through experiments. The improvement point of this paper is to change the traditional kernel correlation filtering algorithm to detect and track at the same time and to combine deep learning with traditional kernel correlation filtering tracking to apply in the process of target tracking; the addition of deep learning network not only can learn more accurate feature representation but also can more effectively cope with the low resolution of video sequences, so that the algorithm in the case of scale mutation achieves more accurate target tracking in the case of scale mutation. To verify the effectiveness of this method in the case of scale mutation, four evaluation criteria, namely, average accuracy, cross-ratio accuracy, temporal robustness, and spatial robustness, are combined to demonstrate the effectiveness of the algorithm in the case of scale mutation. The experimental results verify that the joint detection strategy plays a good role in correcting the tracking drift caused by the subsequent abrupt change of the target scale and the effectiveness of the adaptive template update strategy. By adaptively changing the number of interval frames of neural network redetection to improve the tracking performance, the tracking speed is improved after the fusion of correlation filtering and neural network, and the combination of both is promoted for better application in target tracking tasks.

Target Tracking Algorithm Suitable for Fisheye Camera in Vehicle Environment

Moving target tracking is one of the core issues in the fields of automobile assisted driving and autonomous driving. This algorithm proposes a target tracking algorithm suitable for fisheye cameras in vehicle environment. A background point removal method based on median ordering is proposed to optimize the optical flow method to solve the problem of target scale change in fisheye camera videos. Multi-scale scaling processing for frames, tracking algorithm based on cyclic matrix structure (CSK), target tracking for multiple images after zooming, using fitting to determine the maximum corresponding value, reducing the error caused by target deformation on tracking results. The tracking test results of multiple fisheye camera videos in the vehicle environment show that the algorithm has high real-time tracking accuracy.

Gaussian-response correlation filter for robust visual object tracking

This paper presents a novel correlation filter-based tracking method for robust visual object tracking in the presence of partial occlusion, large-scale variation and model drift. To do this, first, we develop a correlation filter for predicting the target location based on the distribution of correlation response. In this formulation, the correlation response of the target image follows Gaussian distribution to estimate the target location efficiently. Second, the constraints are derived using kernel ridge regression to mitigate the target failure in object tracking. Third, we propose an adaptive scale estimation method to detect the target scale changes during the tracking. In addition, two feature integration is elaborately designed to improve the discriminative strength of the correlation filter. Finally, extensive experimental results on OTB2013, OTB2015, TempleColor128 and UAV123 datasets demonstrate that the proposed method performs favourably against several state-of-the-art methods.