Anchor-free object-detection methods achieve a significant advancement in field of computer vision,particularly in the realm of real-time inferences.However,in remote sensing object detection,anchor-free methods often...Anchor-free object-detection methods achieve a significant advancement in field of computer vision,particularly in the realm of real-time inferences.However,in remote sensing object detection,anchor-free methods often lack of capability in separating the foreground and background.This paper proposes an anchor-free method named probability-enhanced anchor-free detector(ProEnDet)for remote sensing object detection.First,a weighted bidirectional feature pyramid is used for feature extraction.Second,we introduce probability enhancement to strengthen the classification of the object’s foreground and background.The detector uses the logarithm likelihood as the final score to improve the classification of the foreground and background of the object.ProEnDet is verified using the DIOR and NWPU-VHR-10 datasets.The experiment achieved mean average precisions of 61.4 and 69.0 on the DIOR dataset and NWPU-VHR-10 dataset,respectively.ProEnDet achieves a speed of 32.4 FPS on the DIOR dataset,which satisfies the real-time requirements for remote-sensing object detection.展开更多
In anchor-free environments,where no devices with known positions are available,the error growth of autonomous underwater vehicle(AUV)localization and target tracking is unbounded due to the lack of references and the...In anchor-free environments,where no devices with known positions are available,the error growth of autonomous underwater vehicle(AUV)localization and target tracking is unbounded due to the lack of references and the accumulated errors in inertial measurements.This paper aims to improve the localization and tracking accuracy by involving current information as extra references.We first integrate current measurements and maps with belief propagation and design a distributed current-aided message-passing scheme that theoretically solves the localization and tracking problems.Based on this scheme,we propose particle-based cooperative localization and target tracking algorithms,named CaCL and CaTT,respectively.In AUV localization,CaCL uses the current measurements to correct the predicted and transmitted position information and alleviates the impact of the accumulated errors in inertial measurements.With target tracking,the current maps are applied in CaTT to modify the position prediction of the target which is calculated through historical estimates.The effectiveness and robustness of the proposed methods are validated through various simulations by comparisons with alternative methods under different trajectories and current conditions.展开更多
Person Search is a task involving pedestrian detection and person re-identification,aiming to retrieve person images matching a given objective attribute from a large-scale image library.The Person Search models need ...Person Search is a task involving pedestrian detection and person re-identification,aiming to retrieve person images matching a given objective attribute from a large-scale image library.The Person Search models need to understand and capture the detailed features and context information of smaller objects in the image more accurately and comprehensively.The current popular Person Search models,whether end-to-end or two-step,are based on anchor boxes.However,due to the limitations of the anchor itself,the model inevitably has some disadvantages,such as unbalance of positive and negative samples and redundant calculation,which will affect the performance of models.To address the problem of fine-grained understanding of target pedestrians in complex scenes and small sizes,this paper proposes a Deformable-Attention-based Anchor-free Person Search model(DAAPS).Fully Convolutional One-Stage(FCOS),as a classic Anchor-free detector,is chosen as the model’s infrastructure.The DAAPS model is the first to combine the Anchor-free Person Search model with Deformable Attention Mechanism,applied to guide the model adaptively adjust the perceptual.The Deformable Attention Mechanism is used to help the model focus on the critical information and effectively improve the poor accuracy caused by the absence of anchor boxes.The experiment proves the adaptability of the Attention mechanism to the Anchor-free model.Besides,with an improved ResNeXt+network frame,the DAAPS model selects the Triplet-based Online Instance Matching(TOIM)Loss function to achieve a more precise end-to-end Person Search task.Simulation experiments demonstrate that the proposed model has higher accuracy and better robustness than most Person Search models,reaching 95.0%of mean Average Precision(mAP)and 95.6%of Top-1 on the CUHK-SYSU dataset,48.6%of mAP and 84.7%of Top-1 on the Person Re-identification in the Wild(PRW)dataset,respectively.展开更多
针对目前Anchor-free目标检测方法CenterNet(Objects as Points)生成热力图不准确、检测精度不足的问题,提出了一种基于特征迭代聚合的高分辨率表征网络CenterNet-DHRNet。首先,引入高分辨率表征骨干网络,并用迭代聚合的方式对不同分辨...针对目前Anchor-free目标检测方法CenterNet(Objects as Points)生成热力图不准确、检测精度不足的问题,提出了一种基于特征迭代聚合的高分辨率表征网络CenterNet-DHRNet。首先,引入高分辨率表征骨干网络,并用迭代聚合的方式对不同分辨率的特征图进行融合,提高网络的分辨率,有效减少图像在下采样过程中损失的空间语义信息。其次,使用高效通道注意力机制对高分辨率表征骨干网络的输出进行优化。最后,利用结合空洞卷积的空间金字塔池化操作增强网络对不同尺度物体的感受野。实验在PASCAL VOC数据集和KITTI数据集上进行,结果表明:CenterNet-DHRNet精度更高,满足实时检测的性能要求,具有良好的鲁棒性。展开更多
Geographic location of nodes is very useful in a sensor network. Previous localization algorithms assume that there exist some anchor nodes in this kind of network, and then other nodes are estimated to create their c...Geographic location of nodes is very useful in a sensor network. Previous localization algorithms assume that there exist some anchor nodes in this kind of network, and then other nodes are estimated to create their coordinates. Once there are not anchors to be deployed, those localization algorithms will be invalidated. Many papers in this field focus on anchor-based solutions. The use of anchors introduces many limitations, since anchors require external equipments such as global position system, cause additional power consumption. A novel positioning algorithm is proposed to use a virtual coordinate system based on a new concept--virtual anchor. It is executed in a distributed fashion according to the connectivity of a node and the measured distances to its neighbors. Both the adjacent member information and the ranging distance result are combined to generate the estimated position of a network, one of which is independently adopted for localization previously. At the position refinement stage the intermediate estimation of a node begins to be evaluated on its reliability for position mutation; thus the positioning optimization process of the whole network is avoided falling into a local optimal solution. Simulation results prove that the algorithm can resolve the distributed localization problem for anchor-free sensor networks, and is superior to previous methods in terms of its positioning capability under a variety of circumstances.展开更多
The Visual tracking problem can usually be solved in two parts.The first part is to extract the feature of the target and get the candidate region.The second part is to realize the classification of the target and the...The Visual tracking problem can usually be solved in two parts.The first part is to extract the feature of the target and get the candidate region.The second part is to realize the classification of the target and the regression of the bounding box.In recent years,Siameses network in visual tracking problem has always been a frontier research hotspot.In this work,it applies two branches namely search area and tracking template area for similar learning to track.Some related researches prove the feasibility of this network structure.According to the characteristics of two branch shared networks in Siamese network,we also propos a new fully convolutional Siamese network to solve the visual tracking problem.Based on the Siamese network structure,the network we designed adopt a new fusion module,which realizes the fusion of multiple feature layers at different depths.We also devise a better target state estimation criterion.The overall structure is simple,efficient and has wide applicability.We extensive experiments on challenging benchmarks including generic object tracking-10k(GOT-10K),online object tracking benckmark2015(OTB2015)and unmanned air vehicle123(UAV123),and comparisons with state-of-the-art trackers and the fusion module commonly used in the past,Finally,our network performed better under the same backbone,and achieved good tracking effect,which proved the effectiveness and universality of our designed network and feature fusion method.展开更多
由于Anchor-based方法存的一些问题,如目标不规则、手动设计anchor、匹配机制无法匹配极端目标等,提出使用Anchor-free方法用于安防领域的行人与车辆的检测。论文在利用CornerNet-lite进行目标检测的基础上,提出“同类别匹配抑制规则”...由于Anchor-based方法存的一些问题,如目标不规则、手动设计anchor、匹配机制无法匹配极端目标等,提出使用Anchor-free方法用于安防领域的行人与车辆的检测。论文在利用CornerNet-lite进行目标检测的基础上,提出“同类别匹配抑制规则”算法,以降低误报率。所提出的算法基于真实场景的行人与车辆数据集进行测试评估,在不同的场景下,如烈日、阴天、雨雪、夜晚、白天等。实验结果表明,使用安防数据集测试时,改进的算法平均精度(Mean Average Precision,mAP)为0.35,比原方案提高0.05。论文所提出的算法为一般的目标检测算法提供高检测率和低误报率。该算法是有效的,为开发实时行人与车辆检测算法铺平了道路。展开更多
基金supported in part by the National Natural Science Foundation of China(42001408).
文摘Anchor-free object-detection methods achieve a significant advancement in field of computer vision,particularly in the realm of real-time inferences.However,in remote sensing object detection,anchor-free methods often lack of capability in separating the foreground and background.This paper proposes an anchor-free method named probability-enhanced anchor-free detector(ProEnDet)for remote sensing object detection.First,a weighted bidirectional feature pyramid is used for feature extraction.Second,we introduce probability enhancement to strengthen the classification of the object’s foreground and background.The detector uses the logarithm likelihood as the final score to improve the classification of the foreground and background of the object.ProEnDet is verified using the DIOR and NWPU-VHR-10 datasets.The experiment achieved mean average precisions of 61.4 and 69.0 on the DIOR dataset and NWPU-VHR-10 dataset,respectively.ProEnDet achieves a speed of 32.4 FPS on the DIOR dataset,which satisfies the real-time requirements for remote-sensing object detection.
基金supported in part by the National Natural Science Foundation of China(62203299,61773264,61922058,61803261,61801295)the Oceanic Interdisciplinary Program of Shanghai Jiao Tong University(SL2020ZD206,SL2020MS010,SL2020MS015)。
文摘In anchor-free environments,where no devices with known positions are available,the error growth of autonomous underwater vehicle(AUV)localization and target tracking is unbounded due to the lack of references and the accumulated errors in inertial measurements.This paper aims to improve the localization and tracking accuracy by involving current information as extra references.We first integrate current measurements and maps with belief propagation and design a distributed current-aided message-passing scheme that theoretically solves the localization and tracking problems.Based on this scheme,we propose particle-based cooperative localization and target tracking algorithms,named CaCL and CaTT,respectively.In AUV localization,CaCL uses the current measurements to correct the predicted and transmitted position information and alleviates the impact of the accumulated errors in inertial measurements.With target tracking,the current maps are applied in CaTT to modify the position prediction of the target which is calculated through historical estimates.The effectiveness and robustness of the proposed methods are validated through various simulations by comparisons with alternative methods under different trajectories and current conditions.
基金to the Natural Science Foundation of Shanghai under Grant 21ZR1426500,and the Top-Notch Innovative Talent Training Program for Graduate Students of Shanghai Maritime University under Grant 2021YBR008for their generous support and funding through the project funding program.This funding has played a pivotal role in the successful completion of our research.We are deeply appreciative of their invaluable contribution to our research efforts.
文摘Person Search is a task involving pedestrian detection and person re-identification,aiming to retrieve person images matching a given objective attribute from a large-scale image library.The Person Search models need to understand and capture the detailed features and context information of smaller objects in the image more accurately and comprehensively.The current popular Person Search models,whether end-to-end or two-step,are based on anchor boxes.However,due to the limitations of the anchor itself,the model inevitably has some disadvantages,such as unbalance of positive and negative samples and redundant calculation,which will affect the performance of models.To address the problem of fine-grained understanding of target pedestrians in complex scenes and small sizes,this paper proposes a Deformable-Attention-based Anchor-free Person Search model(DAAPS).Fully Convolutional One-Stage(FCOS),as a classic Anchor-free detector,is chosen as the model’s infrastructure.The DAAPS model is the first to combine the Anchor-free Person Search model with Deformable Attention Mechanism,applied to guide the model adaptively adjust the perceptual.The Deformable Attention Mechanism is used to help the model focus on the critical information and effectively improve the poor accuracy caused by the absence of anchor boxes.The experiment proves the adaptability of the Attention mechanism to the Anchor-free model.Besides,with an improved ResNeXt+network frame,the DAAPS model selects the Triplet-based Online Instance Matching(TOIM)Loss function to achieve a more precise end-to-end Person Search task.Simulation experiments demonstrate that the proposed model has higher accuracy and better robustness than most Person Search models,reaching 95.0%of mean Average Precision(mAP)and 95.6%of Top-1 on the CUHK-SYSU dataset,48.6%of mAP and 84.7%of Top-1 on the Person Re-identification in the Wild(PRW)dataset,respectively.
文摘针对目前Anchor-free目标检测方法CenterNet(Objects as Points)生成热力图不准确、检测精度不足的问题,提出了一种基于特征迭代聚合的高分辨率表征网络CenterNet-DHRNet。首先,引入高分辨率表征骨干网络,并用迭代聚合的方式对不同分辨率的特征图进行融合,提高网络的分辨率,有效减少图像在下采样过程中损失的空间语义信息。其次,使用高效通道注意力机制对高分辨率表征骨干网络的输出进行优化。最后,利用结合空洞卷积的空间金字塔池化操作增强网络对不同尺度物体的感受野。实验在PASCAL VOC数据集和KITTI数据集上进行,结果表明:CenterNet-DHRNet精度更高,满足实时检测的性能要求,具有良好的鲁棒性。
基金the National Natural Science Foundation of China (60673054, 60773129)theExcellent Youth Science and Technology Foundation of Anhui Province of China.
文摘Geographic location of nodes is very useful in a sensor network. Previous localization algorithms assume that there exist some anchor nodes in this kind of network, and then other nodes are estimated to create their coordinates. Once there are not anchors to be deployed, those localization algorithms will be invalidated. Many papers in this field focus on anchor-based solutions. The use of anchors introduces many limitations, since anchors require external equipments such as global position system, cause additional power consumption. A novel positioning algorithm is proposed to use a virtual coordinate system based on a new concept--virtual anchor. It is executed in a distributed fashion according to the connectivity of a node and the measured distances to its neighbors. Both the adjacent member information and the ranging distance result are combined to generate the estimated position of a network, one of which is independently adopted for localization previously. At the position refinement stage the intermediate estimation of a node begins to be evaluated on its reliability for position mutation; thus the positioning optimization process of the whole network is avoided falling into a local optimal solution. Simulation results prove that the algorithm can resolve the distributed localization problem for anchor-free sensor networks, and is superior to previous methods in terms of its positioning capability under a variety of circumstances.
文摘The Visual tracking problem can usually be solved in two parts.The first part is to extract the feature of the target and get the candidate region.The second part is to realize the classification of the target and the regression of the bounding box.In recent years,Siameses network in visual tracking problem has always been a frontier research hotspot.In this work,it applies two branches namely search area and tracking template area for similar learning to track.Some related researches prove the feasibility of this network structure.According to the characteristics of two branch shared networks in Siamese network,we also propos a new fully convolutional Siamese network to solve the visual tracking problem.Based on the Siamese network structure,the network we designed adopt a new fusion module,which realizes the fusion of multiple feature layers at different depths.We also devise a better target state estimation criterion.The overall structure is simple,efficient and has wide applicability.We extensive experiments on challenging benchmarks including generic object tracking-10k(GOT-10K),online object tracking benckmark2015(OTB2015)and unmanned air vehicle123(UAV123),and comparisons with state-of-the-art trackers and the fusion module commonly used in the past,Finally,our network performed better under the same backbone,and achieved good tracking effect,which proved the effectiveness and universality of our designed network and feature fusion method.
文摘由于Anchor-based方法存的一些问题,如目标不规则、手动设计anchor、匹配机制无法匹配极端目标等,提出使用Anchor-free方法用于安防领域的行人与车辆的检测。论文在利用CornerNet-lite进行目标检测的基础上,提出“同类别匹配抑制规则”算法,以降低误报率。所提出的算法基于真实场景的行人与车辆数据集进行测试评估,在不同的场景下,如烈日、阴天、雨雪、夜晚、白天等。实验结果表明,使用安防数据集测试时,改进的算法平均精度(Mean Average Precision,mAP)为0.35,比原方案提高0.05。论文所提出的算法为一般的目标检测算法提供高检测率和低误报率。该算法是有效的,为开发实时行人与车辆检测算法铺平了道路。
