基于多尺度注意力网络的行人属性识别算法

为了提高行人属性识别的准确率,提出了一种基于多尺度注意力网络的行人属性识别算法。为了提高算法的特征表达能力和属性判别能力,首先,在残差网络ResNet50的基础上,增加了自顶向下的特征金字塔和注意力模块,自顶向下的特征金字塔由自底向上提取的视觉特征构建;然后,融合特征金字塔中不同尺度的特征,为每层特征的通道注意力赋予不同的权重。最后,改进了模型损失函数以减弱数据不平衡对属性识别率的影响。在RAP和PA-100K数据集上的实验结果表明,与现有算法相比,本算法对行人属性识别的平均精度、准确度、F1性能更好。     

基于行人属性分级识别的行人再识别

为了提高行人再识别算法的识别效果,该文提出一种基于注意力模型的行人属性分级识别神经网络模型,相对于现有算法,该模型有以下3大优点:一是在网络的特征提取部分,设计用于识别行人属性的注意力模型,提取行人属性信息和显著性程度;二是在网络的特征识别部分,针对行人属性的显著性程度和包含的信息量大小,利用注意力模型对属性进行分级识别;三是分析属性之间的相关性,根据上一级的识别结果,调整下一级的识别策略,从而提高小目标属性的识别准确率,进而提高行人再识别的准确率。实验结果表明,该文提出的模型相较于现有方法,有效提高了行人再识别的首位准确率,其中,Market1501数据集上,首位准确率达到了93.1%,在DukeMTMC数据集上,首位准确率达到了81.7%。     

结合时序注意力机制的多特征融合行人序列图像属性识别方法

目前绝大多数的行人属性识别任务都是基于单张图像的,单张图像所含信息有限,而图像序列中包含丰富的有用信息和时序特征,利用序列信息是提高行人属性识别性能的一个重要途径.本文提出了结合时序注意力机制的多特征融合行人序列图像属性识别网络,该网络除了使用常见的空-时二次平均池化特征聚合和空-时平均最大池化特征聚合提取序列的特征外...     

Key points and visible part fusion attention network for occluded pedestrian detection in traffic environments

Aiming at the problem of low detection accuracy of occluded pedestrian in traffic environments, this paper proposes a key points and visible part fusion network for occluded pedestrian detection. The proposed algorithm constructs two attention modules by introducing human key points and the bounding box of visible parts respectively, which suppresses the occluded parts in the channel features and spatial features of pedestrian features respectively. Experimental results on CityPersons and Caltech datasets demonstrate the effectiveness of the proposed algorithm. The missing rate (MR) is reduced to 40.78 on the Heavy subset of the CityPersons dataset and surpasses many outstanding methods.     

Fine-grained-based multi-feature fusion for occluded person re-identification

Many previous occluded person re-identification(re-ID) methods try to use additional clues (pose estimation or semantic parsing models) to focus on non-occluded regions. However, these methods extremely rely on the performance of additional clues and often capture pedestrian features by designing complex modules. In this work, we propose a simple Fine-Grained Multi-Feature Fusion Network (FGMFN) to extract discriminative features, which is a dual-branch structure consisting of global feature branch and partial feature branch. Firstly, we utilize a chunking strategy to extract multi-granularity features to make the pedestrian information contained in it more comprehensive. Secondly, a spatial transformer network is introduced to localize the pedestrian’s upper body, and then introduce a relation-aware attention module to explore the fine-grained information. Finally, we fuse the features obtained from the two branches to obtain a more robust pedestrian representation. Extensive experiments verify the effectiveness of our method under the occlusion scenario.     

改进YOLOv4算法的复杂视觉场景行人检测方法

复杂视觉场景下存在过暗或者过曝的光照、恶劣的天气、严重遮挡、行人尺寸差别大以及图像模糊等问题,大大增加了行人检测的难度。因此,针对复杂视觉场景下行人检测准确度低、漏检严重的问题,提出了改进的YOLOv4算法以增强复杂视觉场景下的行人检测效果。首先,构建复杂视觉场景下的行人数据集。然后,在主干网中加入混合空洞卷积,提高网络对行人特征的提取能力。最后,提出空间锯齿空洞卷积结构,代替空间金字塔池化结构,获取更多细节特征。实验表明,在本文构建的行人数据集上,改进后的 YOLOv4算法的平均精度（average precision,AP）达到了90.08%,相比原YOLOv4算法提高了7.2%,对数平均漏检率（log-average miss rate,LAMR）降低了13.69%。     

基于特征增强模块的小尺度行人检测

行人检测中,小尺度行人时常被漏检、误检。为了提升小尺度行人的检测准确率并且降低其漏检率,该文提出一个特征增强模块。首先,考虑到小尺度行人随着网络加深特征逐渐减少的问题,特征融合策略突破特征金字塔层级结构的约束,融合深层、浅层特征图,保留了大量小尺度行人特征。然后,考虑到小尺度行人特征容易与背景信息发生混淆的问题,通过自注意力模块联合通道注意力模块建模特征图空间、通道关联性,利用小尺度行人上下文信息和通道信息,增强了小尺度行人特征并且抑制了背景信息。最后,基于特征增强模块构建了一个小尺度行人检测器。所提方法在CrowdHuman数据集中小尺度行人的检测准确率为19.8%,检测速度为22帧/s,在CityPersons数据集中小尺度行人的误检率为13.1%。结果表明该方法对于小尺度行人的检测效果优于其他对比算法且实现了较快的检测速度。     

融合多尺度信息和跨维特征引导的轻量行人检测

针对复杂道路场景下行人检测精度与速度难以提升的问题,提出一种融合多尺度信息和跨维特征引导的轻量级行人检测算法。首先以高性能检测器YOLOX为基础框架,构建多尺度轻量卷积并嵌入主干网络中,以获取多尺度特征信息。然后设计了一种端到端的轻量特征引导注意力模块,采用跨维通道加权的方式将空间信息与通道信息融合,引导模型关注行人的可视区域。最后为减少模型在轻量化过程中特征信息的损失,使用增大感受野的深度可分离卷积构建特征融合网络。实验结果表明,相比于其他主流检测算法,所提算法在KITTI数据集上达到了71.03%的检测精度和80 FPS的检测速度,在背景复杂、密集遮挡、尺度不一等场景中都具有较好的鲁棒性和实时性。     

基于多样化局部注意力网络的行人重识别

针对现实场景中行人图像被遮挡以及行人姿态或视角变化造成的未对齐问题,该文提出一种基于多样化局部注意力网络(DLAN)的行人重识别(Re-ID)方法.首先,在骨干网络后分别设计了全局网络和多分支局部注意力网络,一方面学习全局的人体空间结构特征,另一方面自适应地获取人体不同部位的显著性局部特征;然后,构造了一致性激活惩罚函...     

结合一阶和二阶空间信息的行人重识别

针对行人重识别中行人检测误差引起的空间错位,基于局部的深度网络模型仅学习相邻局部关系,导致远距离局部相关性缺失,因此,提出了一种结合一阶和二阶空间信息的行人重识别算法。在主干网络上,学习一阶空间掩模对输入图像的空间权值进行微调,以减少背景干扰;通过二阶空间掩模对远距离的依赖关系进行建模,并将局部特征集成到依赖模型中,以获取全局特征表示。局部分支引入DropBlock对抽取的行人特征进行正则化,避免了网络模型过于依赖特定部位特征。训练阶段用标签平滑分类损失和引入正样本中心的三元组损失联合优化整个网络。在Market-1501和DukeMTMC-reID数据集上的实验结果表明,相比其他主流算法,本算法的行人重识别精度更高,且提取的行人特征判别性和鲁棒性更好。     

Vehicle and pedestrian detection method based on improved YOLOv4-tiny

Aiming at the problem of low detection accuracy of vehicle and pedestrian detection models, this paper proposes an improved you only look once v4 (YOLOv4)-tiny vehicle and pedestrian target detection algorithm. Convolutional block attention module (CBAM) is introduced into cross stage partial Darknet-53 (CSPDarknet53)-tiny module to enhance feature extraction capabilities. In addition, the cross stage partial dense block layer (CSP-DBL) module is used to replace the original simple convolutional module superposition, which compensates for the high-resolution characteristic information and further improves the detection accuracy of the network. Finally, the test results on the BDD100K traffic dataset show that the mean average precision (mAP) value of the final network of the proposed method is 88.74%, and the detection speed reaches 63 frames per second (FPS), which improves the detection accuracy of the network and meets the real-time detection speed.     

Learning an attention-aware parallel sharing network for facial attribute recognition

Existing multi-task learning based facial attribute recognition (FAR) methods usually employ the serial sharing network, where the high-level global features are used for attribute prediction. However, the shared low-level features with valuable spatial information are not well exploited for multiple tasks. This paper proposes a novel Attention-aware Parallel Sharing network termed APS for effective FAR. To make full use of the shared low-level features, the task-specific sub-networks can adaptively extract important features from each block of the shared sub-network. Furthermore, an effective attention mechanism with multi-feature soft-alignment modules is employed to evaluate the compatibility of the local and global features from the different network levels for discriminating attributes. In addition, an adaptive Focal loss penalty scheme is developed to automatically assign weights to handle the problems of class imbalance and hard example mining for FAR. Experiments demonstrate that the proposed method achieves better performance than the state-of-the-art FAR methods.     

Person re-identification based on improved attention mechanism and global pooling method

Deep network has become a new favorite for person re-identification (Re-ID), whose research focus is how to effectively extract the discriminative feature representation for pedestrians. In the paper, we propose a novel Re-ID network named as improved ReIDNet (iReIDNet), which can effectively extract the local and global multi-granular feature representations of pedestrians by a well-designed spatial feature transform and coordinate attention (SFTCA) mechanism together with improved global pooling (IGP) method. SFTCA utilizes channel adaptability and spatial location to infer a 2D attention map and can help iReIDNet to focus on the salient information contained in pedestrian images. IGP makes iReIDNet capture more effectively the global information of the whole human body. Besides, to boost the recognition accuracy, we develop a weighted joint loss to guide the training of iReIDNet. Comprehensive experiments demonstrate the availability and superiority of iReIDNet over other Re-ID methods. The code is available at https://github.com/XuRuyu66/ iReIDNet.     

引入语义部位约束的行人再识别

为减轻行人图片中的背景干扰,使网络着重于行人前景并且提高前景中人体部位的利用率,该文提出引入语义部位约束(SPC)的行人再识别网络。在训练阶段,首先将行人图片同时输入主干网络和语义部位分割网络,分别得到行人特征图和部位分割图;然后,将部位分割图与行人特征图融合,得到语义部位特征;接着,对行人特征图进行池化得到全局特征;最后,同时使用身份约束和语义部位约束训练网络。在测试阶段,由于语义部位约束使得全局特征拥有部位信息,因此测试时仅使用主干网络提取行人的全局信息即可。在大规模公开数据集上的实验结果表明,语义部位约束能有效使得网络提高辨别行人身份的能力并且缩减推断网络的计算花费。与现有方法比较,该文网络能更好地抵抗背景干扰,提高行人再识别性能。     

基于多尺度和注意力融合学习的行人重识别

行人重识别的关键依赖于行人特征的提取,卷积神经网络具有强大的特征提取以及表达能力。针对不同尺度下可以观察到不同的特征,该文提出一种基于多尺度和注意力网络融合的行人重识别方法(MSAN)。该方法通过对网络不同深度的特征进行采样,将采样的特征融合后对行人进行预测。不同深度的特征图具有不同的表达能力,使网络可以学习到行人身上更加细粒度的特征。同时将注意力模块嵌入到残差网络中,使得网络能更加关注于一些关键信息,增强网络特征学习能力。所提方法在Market1501, DukeMTMC-reID和MSMT17_V1数据集上首位准确率分别到了95.3%, 89.8%和82.2%。实验表明,该方法充分利用了网络不同深度的信息和关注的关键信息,使模型具有很强的判别能力,而且所提模型的平均准确率优于大多数先进算法。     

3D pedestrian localization fusing via monocular camera

In the field of intelligent transportation, autonomous driving technologies, especially visual sensing solutions have attracted increasing attention in recent years. There are still some challenges in pedestrian location based on the monocular camera, as the pedestrian is a non-rigid object and its depth information cannot be obtained from the monocular camera easily and accurately. In this paper, a pedestrian location framework based on monocular cameras is proposed. The framework consists of three parts: coarse positioning, auxiliary information generation and information fusion. In the part of coarse positioning, the human skeleton information is obtained from the monocular images and a light-weight feed-forward neural network is used to predict the pedestrian position based on the skeleton information. In the part of auxiliary information generation, pseudo-LiDAR points with pedestrian depth information are generated from the monocular images through an auxiliary network. Finally, the outputs of the above two parts are fused to achieve the pedestrian location. The experimental results on KITTI dataset show that our method has achieved better performance than other methods.     

基于双阶段轻量YOLO的红外行人伪影检测算法北大核心CSCD

由于镜面回波效应,红外图像采集过程中行人不可避免出现反射倒影(本文简称“伪影”)区域,此时对后续行人检测会造成一定程度影响,对上述情况,本文提出一种基于轻量YOLO(You Only Look Once)的双阶段网络检测框架,先检测“行人-伪影”联合区域再精准定位伪影位置。首先,针对YOLOv5s轻量检测算法进行改进,使用LSM(Light Sample Module)双分支结构替换原下采样部分,并嵌入注意力机制来提高模型的特征整合能力,实现红外图像的背景过滤和联合区域提取。其次,对联合区域进行无失真矩形填充保持原始特征,设计轻量级行人伪影定位网络LS YOLO(Light Structur YOLO)检测联合区域获得最终的伪影位置坐标。实验结果表明,本文算法能够满足实时检测要求,在数据集中,相比其他算法获得更好的检测效果,行人伪影的检测正确率达到9545%。     

基于二阶混合注意力的行人重识别

基于注意力机制的行人重识别方法更多利用图像中 一阶信息,忽略了特征中二阶信息 ,不能挖掘特征图之间的相关性和细粒度信息。提出一种基于二阶混合注意力的行人重 识别算法(second-order mixed attention module,SMAN)。二阶混合注意力模块(second-order mixed attention module,SOMA)由二阶通道注意力(second-order channel attention,SOCA)和二阶空间注意力模 块(second-order information,SOSA)组成,该方法将全局协方差池函数嵌入到SOCA和SOSA模块中,学习特征中二阶信息 。SOCA模块学习特征图之间相关性,SOSA模块则重新为特征图分配权重,关注特征图空间域 的细粒度信息。SMAN算法在Market-1501和 DukeMTMC-ReID数据集上的首位准确率分别 为 94.3%和87.1%,mAP分别达到85.7%和74.5%,同时使用类激活图验证SOMA模块的影响 ,实验表 明SMAN算法充分利用特征图的通道域和空间域中二阶信息。算法的性能优于现有的一些基于 注意力机制行人重识别方法,甚至接近某些优秀的方法。     

基于YOLO的轻量红外图像行人检测方法

红外图像中行人的快速检测一直是计算机视觉领域的热点和难点。针对红外图像行人目标检测算法检测速度和检测精度难以平衡,算法模型体积较大,在中低性能设备中难以部署和实时运行的问题,提出了一种基于YOLO算法的轻量红外图像行人检测方法。在分析了MobileNet v3等轻量网络在YOLO v3算法上的性能和特点之后,该方法提出了引入注意力机制的轻量特征提取网络(CSPmini a)、特征融合模块和解耦检测端分类回归结构三种改进措施,在满足网络模型轻量的情况下保证了一定的检测精度。实验表明,该方法有效的实现了红外图像行人目标检测的准确性和快速性。     

Internal and external spatial–temporal constraints for person reidentification

Spatial–temporal information is easy to achieve in a practical surveillance scene, but it is often neglected in most current person reidentification (ReID) methods. Employing spatial–temporal information as a constrain has been verified as beneficial for ReID. However, there is no effective modeling according to the pedestrian movement law. In this paper, we present a ReID framework with internal and external spatial–temporal constraints, termed as IESC-ReID. A novel residual spatial attention module is proposed to build a spatial–temporal constraint and increase the robustness to partial occlusions or camera viewpoint changes. A Laplace-based spatial–temporal constraint is also introduced to eliminate irrelevant gallery images, which are gathered by the internal learning network. IESC-ReID constrains the attention within the functioning range of the channel space, and utilizes additional spatial–temporal constrains to further constrain results. Intensive experiments show that these constraints consistently improve the performance. Extensive experimental results on numerous publicly available datasets show that the proposed method outperforms several state-of-the-art ReID algorithms. Our code is publicly available at https://github.com/jiaming-wang/IESC.