目的在行人再识别中,行人朝向变化会导致表观变化,进而导致关联错误。现有方法通过朝向表示学习和基于朝向的损失函数来改善这一问题。然而,大多数朝向表示学习方法主要以嵌入朝向标签为主,并没有显式的向模型传达行人姿态的空间结构,...目的在行人再识别中,行人朝向变化会导致表观变化,进而导致关联错误。现有方法通过朝向表示学习和基于朝向的损失函数来改善这一问题。然而,大多数朝向表示学习方法主要以嵌入朝向标签为主,并没有显式的向模型传达行人姿态的空间结构,从而减弱了模型对朝向的感知能力。此外,基于朝向的损失函数通常对相同身份的行人进行朝向聚类,忽略了由表观相似且朝向相同的负样本造成的错误关联的问题。方法为了应对这些挑战,提出了面向行人再识别的朝向感知特征学习。首先,提出了基于人体姿态的朝向特征学习,它能够显式地捕捉人体姿态的空间结构。其次,提出的朝向自适应的三元组损失主动增大表观相似且相同朝向行人之间的间隔,进而将它们分离。结果本文方法在大规模的行人再识别公开数据集MSMT17(multi-scene multi-time person ReID dataset)、Market1501等上进行测试。其中,在MSMT17数据集上,相比于性能第2的UniHCP(unified model for human-centric perceptions)模型,Rank1和mAP值分别提高了1.7%和1.3%;同时,在MSMT17数据集上的消融实验结果证明本文提出的算法有效改善了行人再识别的关联效果。结论本文方法能够有效处理上述挑战导致的行人再识别系统中关联效果变差的问题。展开更多
现有行人重识别技术主要关注水平视角下的图像。在例如无人超市这类特定场景下,摄像头以俯视角度拍摄,仅能获得有限的行人信息。针对此问题,将多模态视觉Transformer应用于俯视图行人重识别任务,利用俯视数据集中额外的深度模态来提高...现有行人重识别技术主要关注水平视角下的图像。在例如无人超市这类特定场景下,摄像头以俯视角度拍摄,仅能获得有限的行人信息。针对此问题,将多模态视觉Transformer应用于俯视图行人重识别任务,利用俯视数据集中额外的深度模态来提高俯视图的检索精度。具体而言,提出一种基于RGB(red,green,blue)与深度多模态视觉Transformer的特征提取方法,利用双流网络提取数据集的深度信息,自集成多个自注意力层的特征输出,以此作为最终的图像特征,并对损失函数进行改进,从而提高了模型的检索效果。通过在俯视图数据集TVPR(top-view person re-identification)和TVPR2上开展实验,结果表明:所提方法能有效提升检索效果,且超过了几种先进的俯视图行人重识别方法。展开更多
This paper presents a travel-aid system for sight-impaired pedestrians in outdoor environment.Unlike many existing travel-aid systems that rely on stereo-vision,the proposed system aims to detect generic obstacles in ...This paper presents a travel-aid system for sight-impaired pedestrians in outdoor environment.Unlike many existing travel-aid systems that rely on stereo-vision,the proposed system aims to detect generic obstacles in a cluttered road environment by using just the single camera attached to user's body.To achieve this goal,the original image is re-sampled and mapped to a bird-eye view virtual plane,on which background edges are sub-sampled while obstacle edges are oversampled.Morphology filters with connected component analysis are used to enhance obstacle edges as edge-blobs with larger size,whereas sparse edges from background are filtered out.Based on the identified obstacles,safe walking areas are estimated by calculating a polar edge histogram.The algorithm is tested in different sidewalk scenes with complex pavements,and its efficiency has been confirmed.展开更多
We study the influence of limited-view scanning on the depth imaging of photoacoustic tomography. The situation, in which absorbers are located at different depths with respect to the limited-view scanning trajectory,...We study the influence of limited-view scanning on the depth imaging of photoacoustic tomography. The situation, in which absorbers are located at different depths with respect to the limited-view scanning trajectory, is called depth imaging and is investigated in this paper. The results show that limited-view scanning causes the reconstructed intensity of deep absorbers to be weaker than that of shallow ones and that deep absorbers will be invisible if the scanning range is too small. The concept of effective scanning angle is proposed to analyse that phenomenon. We find that an effective scanning angle can well predict the relationship between scanning angle and the intensity ratio of absorbers. In addition, limited-view scanning is employed to improve image quality.展开更多
文摘目的在行人再识别中,行人朝向变化会导致表观变化,进而导致关联错误。现有方法通过朝向表示学习和基于朝向的损失函数来改善这一问题。然而,大多数朝向表示学习方法主要以嵌入朝向标签为主,并没有显式的向模型传达行人姿态的空间结构,从而减弱了模型对朝向的感知能力。此外,基于朝向的损失函数通常对相同身份的行人进行朝向聚类,忽略了由表观相似且朝向相同的负样本造成的错误关联的问题。方法为了应对这些挑战,提出了面向行人再识别的朝向感知特征学习。首先,提出了基于人体姿态的朝向特征学习,它能够显式地捕捉人体姿态的空间结构。其次,提出的朝向自适应的三元组损失主动增大表观相似且相同朝向行人之间的间隔,进而将它们分离。结果本文方法在大规模的行人再识别公开数据集MSMT17(multi-scene multi-time person ReID dataset)、Market1501等上进行测试。其中,在MSMT17数据集上,相比于性能第2的UniHCP(unified model for human-centric perceptions)模型,Rank1和mAP值分别提高了1.7%和1.3%;同时,在MSMT17数据集上的消融实验结果证明本文提出的算法有效改善了行人再识别的关联效果。结论本文方法能够有效处理上述挑战导致的行人再识别系统中关联效果变差的问题。
文摘现有行人重识别技术主要关注水平视角下的图像。在例如无人超市这类特定场景下,摄像头以俯视角度拍摄,仅能获得有限的行人信息。针对此问题,将多模态视觉Transformer应用于俯视图行人重识别任务,利用俯视数据集中额外的深度模态来提高俯视图的检索精度。具体而言,提出一种基于RGB(red,green,blue)与深度多模态视觉Transformer的特征提取方法,利用双流网络提取数据集的深度信息,自集成多个自注意力层的特征输出,以此作为最终的图像特征,并对损失函数进行改进,从而提高了模型的检索效果。通过在俯视图数据集TVPR(top-view person re-identification)和TVPR2上开展实验,结果表明:所提方法能有效提升检索效果,且超过了几种先进的俯视图行人重识别方法。
基金The MKE(The Ministry of Knowledge Economy),Korea,under the ITRC(Infor mation Technology Research Center)support programsupervised by the NIPA(National IT Industry Promotion Agency)(NIPA-2012-(C1090-1021-0010))The Brain Korea 21 Project in 2012
文摘This paper presents a travel-aid system for sight-impaired pedestrians in outdoor environment.Unlike many existing travel-aid systems that rely on stereo-vision,the proposed system aims to detect generic obstacles in a cluttered road environment by using just the single camera attached to user's body.To achieve this goal,the original image is re-sampled and mapped to a bird-eye view virtual plane,on which background edges are sub-sampled while obstacle edges are oversampled.Morphology filters with connected component analysis are used to enhance obstacle edges as edge-blobs with larger size,whereas sparse edges from background are filtered out.Based on the identified obstacles,safe walking areas are estimated by calculating a polar edge histogram.The algorithm is tested in different sidewalk scenes with complex pavements,and its efficiency has been confirmed.
基金Project supported by the National Basic Research Program of China(Grant No.2012CB921504)the National Natural Science Foundation of China(Grant Nos.10874088,10904069,and 11028408)the Natural Science Foundation of Jiangsu Province,China(Grant No.SBK201021985)
文摘We study the influence of limited-view scanning on the depth imaging of photoacoustic tomography. The situation, in which absorbers are located at different depths with respect to the limited-view scanning trajectory, is called depth imaging and is investigated in this paper. The results show that limited-view scanning causes the reconstructed intensity of deep absorbers to be weaker than that of shallow ones and that deep absorbers will be invisible if the scanning range is too small. The concept of effective scanning angle is proposed to analyse that phenomenon. We find that an effective scanning angle can well predict the relationship between scanning angle and the intensity ratio of absorbers. In addition, limited-view scanning is employed to improve image quality.
