基于改进Yolov5s的混凝土坝裂缝检测研究

大坝作为我国的重要水利工程，其安全性不言而喻。裂缝作为大坝的主要安全隐患，大坝裂缝的智能化检测，对管控大坝风险具有非凡的意义。提出一种基于改进YOLOv5s目标检测算法的大坝裂缝检测方法。首先，在YOLOv5s的主干特征提取网络中引入轻量级GhostNet中的Ghost模块，对YOLOv5s的主干网络进行优化，得到轻量级的模型YOLOv5s-Ghost,以降低模型的复杂度，提高裂缝的检测速度；然后在模型预测输出端融合高效的CA(Coordinate Attention)注意力机制进一步增强裂缝特征提取能力，提高裂缝模型检测的性能。实验结果表明：该方法与现有的YOLOv5s相比，模型大小复杂度降低了44.8%,准确率提升了2.6%,验证了改进方案的有效性，提高了裂缝检测效率。通过引入GhostNet中的Ghost模块和融合CA注意力机制相结合，使得YOLOv5模型的复杂度降低，参数量减少，实现了对裂缝检测速度与精度的提高，增强了网络性能。     

安全帽与反光衣的轻量化检测：改进YOLOv5s的算法

安全帽与反光衣检测对生产与交通环境的安全管理具有重要意义。针对目前安全帽和反光衣检测算法参数量大、计算量大和模型体积较大等问题，提出了一种基于YOLOv5s轻量化改进的检测算法。引入GhostNet网络结构中的Ghost模块代替原有的部分卷积与C3模块，大大降低了模型的复杂度。在主干网中增加CA注意力机制，抑制无效信息，增强对特征丰富区域的提取。用C3CBAM代替neck层的C3模块，既减少参数量，又提高了检测精度。实验结果表明，改进模型的mAP（平均精度）为93.6%，参数量为4.28×106，计算量为9.2 GFLOPs，模型大小为8.58 MB。与YOLOv5模型相比较参数量减少了39%，计算量减少了41.7%，模型大小降低37.3%。该检测算法既保证了检测的识别准确率，又实现了检测算法的轻量化。     

引入Ghost模块和ECA的YOLOv4公路路面裂缝检测方法

针对目前公路路面裂缝种类和尺度多样导致路面病害检测困难的问题,提出一种基于GhostNet的轻量化无人机图像裂缝检测方法检测不同种类路面裂缝。首先,引入轻量级GhostNet中的Ghost模块优化YOLOv4主干特征提取网络,得到轻量化模型YOLOv4-Light,以降低模型复杂度,并提高裂缝检测速度;然后,在模型预测输出端融合高效通道注意力（ECA）机制,从而进一步增强裂缝特征提取能力,提高裂缝检测精度。仿真实验结果表明,所提方法与现有的YOLOv4相比,模型大小降低了82.31%,模型参数量减少了82.56%,并提高了裂缝检测效率,能够满足公路运输过程中出现的不同类型的裂缝检测需求。     

基于轻量化神经网络的社交距离检测

保持安全社交距离是有效防止病毒传播的重要手段之一,不仅可以减少感染者数量和医疗负担,同时也极大降低死亡率.在YOLOv4框架基础上使用轻量化网络E-GhostNet代替原网络中的CSPDarknet-53, E-GhostNet网络在输入数据和原始Ghost模块生成的输出特征之间建立关系,使网络能够捕获上下文特征.然后,在E-Ghost Net中引入坐标注意力机制(CA)增强模型对有效特征的关注.另外,使用SIoU损失函数更换CIoU损失获得更快的收敛速度和优化效果.最后,结合DeepSORT多目标跟踪算法来检测和标记行人,并使用仿射变换(IPM)判定行人间距离的违规行为.实验结果显示,该网络检测速度为40 FPS,精度值达到85.71%,相比原始GhostNet算法提升2.57%,达到实时行人距离检测的效果.     

改进YOLOv4的轻量级遥感图像建筑物检测模型

针对现有建筑物检测模型检测精度低下,模型体积较大,导致遥感图像检测速度和精度无法平衡,不利于后期部署等问题,提出一种基于YOLOv4优化的轻量级遥感图像建筑物检测模型。利用轻量化网络GhostNet替换CSP DarkNet53进行特征提取;借鉴稠密连接思想,提出了Dense-PANet特征融合模块;将ECA注意力机制引入Ghost模块,替换特征融合颈部网络的传统卷积。实验结果表明,提出的模型与YOLOv4相比,牺牲少量检测速度,但是平均精度提高了0.96个百分点,召回率提升了1.08个百分点,模型体积降低了71.39%,浮点计算量降低了76.60%,能有效满足遥感图像建筑物检测的需求。     

基于混合注意力机制的目标跟踪算法

针对全卷积孪生网络目标跟踪算法(SiamFC)在目标形变、遮挡和快速运动等复杂场景中易导致跟踪失败的问题,提出一种利用混合注意力机制增强网络识别能力的算法(SiamMA)。首先,在训练阶段提出堆叠裁剪法构建自对抗训练样本对,以模拟实际跟踪时的复杂场景,使训练的网络模型具有更强的泛化性。其次提出混合注意力机制,在网络不同分支融合使用空间注意力和通道注意力网络模块,有效抑制了特征图中的背景干扰,提升算法的鲁棒性。采用GOT-10k、UAV123等4种数据集进行算法性能评测实验,结果表明本文算法的跟踪成功率、精度等主要性能指标较SiamFC和KCF等6种经典算法均有提升,平均速度达到60 fps。     

基于GhostNet与注意力机制的YOLOv5交通目标检测

针对交通目标检测模型参数量大、检测精度低、检测速度慢、泛化性差等问题,提出一种基于GhostNet与注意力机制的YOLOv5交通目标实时检测模型.采用基于遗传算法的K-means聚类方法获取适用于车辆检测的最佳预选框;采用轻量的Ghost卷积提取目标特征,并构建基于CSP结构的C3Ghost模块,大幅度压缩模型参数量,降低计算成本,提高计算速度;在特征融合层添加Transformer block和CBAM注意力模块,来探索模型特征提取潜力以及为模型在密集对象的场景中寻找注意力区域; UA-DETRAC数据集上的消融实验和综合性能评价结果表明所提模型平均精度达到98.68%,参数量为47 M,检测速度为65 FPS,与YOLOv5相比,参数量压缩了34%,速度提升43%,平均精度提高了1.05%.     

改进YOLOv7与DeepSORT的佩戴口罩行人跟踪

针对视频序列中因脸部遮挡、漏检而造成的无法正确判断行人是否佩戴口罩的问题,提出一种基于改进YOLOv7与DeepSORT的佩戴口罩行人跟踪算法。该算法将口罩检测、行人检测与跟踪相结合,通过在YOLOv7的主干网络中添加注意力机制,增加浅层特征图,加强网络对小目标的感知能力,提高口罩检测与行人检测精度;帧内关系模块利用匈牙利算法进行帧内目标关联,对行人进行口罩佩戴标记;将方向差因素加入到DeepSORT算法的关联代价中,消除跟踪轨迹的历史预测方向和新检测速度方向不一致问题;使用改进的DeepSORT算法对行人进行跟踪,并对每条轨迹进行口罩佩戴标记更新,实现对佩戴口罩与未佩戴口罩行人的跟踪。实验结果表明,改进的YOLOv7网络平均检测精度mAP50相比原始算法提升了3.83个百分点;在MOT16数据集上,该算法的跟踪准确性MOTA相较DeepSORT算法提高了17.1个百分点,跟踪精度MOTP提高了2.6个百分点。与检测算法相比,提出的算法能够跟踪到更多的行人是否佩戴了口罩,具有更好的效果。     

基于孪生网络和Transformer的小目标跟踪算法SiamTrans

针对现有小目标跟踪算法的鲁棒性差、精度及成功率低的问题，提出一种基于孪生网络和Transformer的小目标跟踪算法SiamTrans。首先，基于Transformer机制设计一种相似度响应图计算模块。该模块叠加若干层特征编码-解码结构，并利用多头自注意力机制和多头跨注意力机制在不同层次的搜索区域特征图中查询模板特征图信息，从而避免陷入局部最优解，并获得一个高质量的相似度响应图；其次，在预测子网中设计一个基于Transformer机制的预测模块（PM），并利用自注意力机制处理预测分支特征图中的冗余特征信息，以提高不同预测分支的预测精度。在Small90数据集上，相较于TransT(Transformer Tracking)算法，所提算法的跟踪精度和跟踪成功率分别高8.0和9.5个百分点。可见，所提出的算法具有更优异的小目标跟踪性能。     

融合残差连接与通道注意力机制的Siamese目标跟踪算法

针对Siamese跟踪算法在目标形变、相似物体干扰等复杂情况下容易跟踪漂移或丢失的问题,提出一种融合残差连接与通道注意力机制的目标跟踪算法.首先,通过残差连接将模板分支网络提取的浅层结构特征与深层语义特征进行有效的融合,以提高模型的表征能力;其次,引入通道注意力模块,使模型自适应地对不同语义目标特征通道加权,以提高模型...     

Probabilistic people tracking with appearance models and occlusion classification: The AD-HOC system

AD-HOC (Appearance Driven Human tracking with Occlusion Classification) is a complete framework for multiple people tracking in video surveillance applications in presence of large occlusions. The appearance-based approach allows the estimation of the pixel-wise shape of each tracked person even during the occlusion. This peculiarity can be very useful for higher level processes, such as action recognition or event detection. A first step predicts the position of all the objects in the new frame while a MAP framework provides a solution for best placement. A second step associates each candidate foreground pixel to an object according to mutual object position and color similarity. A novel definition of non-visible regions accounts for the parts of the objects that are not detected in the current frame, classifying them as dynamic, scene or apparent occlusions. Results on surveillance videos are reported, using in-house produced videos and the PETS2006 test set.     

Visual object tracking—classical and contemporary approaches

Visual object tracking (VOT) is an important subfield of computer vision. It has widespread application domains, and has been considered as an important part of surveillance and security system. VOA facilitates finding the position of target in image coordinates of video frames.While doing this, VOA also faces many challenges such as noise, clutter, occlusion, rapid change in object appearances, highly maneuvered (complex) object motion, illumination changes. In recent years, VOT has made significant progress due to availability of low-cost high-quality video cameras as well as fast computational resources, and many modern techniques have been proposed to handle the challenges faced by VOT. This article introduces the readers to 1) VOT and its applications in other domains, 2) different issues which arise in it, 3) various classical as well as contemporary approaches for object tracking, 4) evaluation methodologies for VOT, and 5) online resources, i.e., annotated datasets and source code available for various tracking techniques.     

Real time 3D face and facial feature tracking

Detecting and tracking human faces in video sequences is useful in a number of applications such as gesture recognition and human-machine interaction. In this paper, we show that online appearance models (holistic approaches) can be used for simultaneously tracking the head, the lips, the eyebrows, and the eyelids in monocular video sequences. Unlike previous approaches to eyelid tracking, we show that the online appearance models can be used for this purpose. Neither color information nor intensity edges are used by our proposed approach. More precisely, we show how the classical appearance-based trackers can be upgraded in order to deal with fast eyelid movements. The proposed eyelid tracking is made robust by avoiding eye feature extraction. Experiments on real videos show the usefulness of the proposed tracking schemes as well as their enhancement to our previous approach.

单目标跟踪算法的研究与分析

运动目标跟踪是计算机视觉的一个中心研究问题,为视频内容的理解提供重要的信息。首先介绍了目标跟踪的国内外研究现状,重点归纳分析了运动目标跟踪方法的分类及其发展过程中的提出的各种算法,对其关键技术进行了剖析和比较。     

Incremental Focus of Attention for Robust Vision-Based Tracking

We present the Incremental Focus of Attention (IFA) architecture for robust, adaptive, real-time motion tracking. IFA systems combine several visual search and vision-based tracking algorithms into a layered hierarchy. The architecture controls the transitions between layers and executes algorithms appropriate to the visual environment at hand: When conditions are good, tracking is accurate and precise; as conditions deteriorate, more robust, yet less accurate algorithms take over; when tracking is lost altogether, layers cooperate to perform a rapid search for the target and continue tracking.Implemented IFA systems are extremely robust to most common types of temporary visual disturbances. They resist minor visual perturbances and recover quickly after full occlusions, illumination changes, major distractions, and target disappearances. Analysis of the algorithm's recovery times are supported by simulation results and experiments on real data. In particular, examples show that recovery times after lost tracking depend primarily on the number of objects visually similar to the target in the field of view.     

Mean-shift跟踪中的状态分析及处理

在Mean-shift算法中,对跟踪状态的误判会引起错误的模板调整策略,造成目标丢失。提出了一种新的跟踪状态分析方法,该算法首先通过分析目标特征与背景特征的相对关系,引入特征增强函数,并在此基础上构造了新的背景模板。然后通过对跟踪中各模板相似系数动态变化过程的综合分析,准确判定目标所处跟踪状态,以采取相应的模板调整策略。实验表明,该算法可以有效增强Mean-shift算法在复杂条件下的跟踪效果,具有较好的稳健性。     

数字化封闭图形对象的面积计算

四方向跟踪算法可以用于计算数字化封闭图形对象的面积,但由于四方向跟踪算法有其自身的不足之处,因此笔者对其作了改进,得到四方向试探跟踪算法和八方向跟踪算法。本文主要是为改进后的算法给出与四方向跟踪算法类似的计算图形对象面积的方法。     

视觉跟踪技术研究现状及其展望*

对运动目标进行视觉跟踪是机器人技术、计算机视觉等领域研究的热点之一。首先就跟踪技术的研究历史和现状进行回顾和分析,论述按照基于目标区域的跟踪、基于目标特征的跟踪、基于变形模板的跟踪和基于目标模型的跟踪四个分支分别展开;其次,对人们在该领域研究所面临的难点进行了归纳和总结;最后对分形理论和生物视觉在跟踪策略中的应用前景进行了展望。     

Probabilistic Tracking with Exemplars in a Metric Space

A new, exemplar-based, probabilistic paradigm for visual tracking is presented. Probabilistic mechanisms are attractive because they handle fusion of information, especially temporal fusion, in a principled manner. Exemplars are selected representatives of raw training data, used here to represent probabilistic mixture distributions of object configurations. Their use avoids tedious hand-construction of object models, and problems with changes of topology.Using exemplars in place of a parameterized model poses several challenges, addressed here with what we call the Metric Mixture (M²) approach, which has a number of attractions. Principally, it provides alternatives to standard learning algorithms by allowing the use of metrics that are not embedded in a vector space. Secondly, it uses a noise model that is learned from training data. Lastly, it eliminates any need for an assumption of probabilistic pixelwise independence.Experiments demonstrate the effectiveness of the M² model in two domains: tracking walking people using chamfer distances on binary edge images, and tracking mouth movements by means of a shuffle distance.     

Learning-based dynamic coupling of discrete and continuous trackers

We present a data-driven dynamic coupling between discrete and continuous methods for tracking objects of high dofs, which overcomes the limitations of previous techniques. In our approach, two trackers work in parallel, and the coupling between them is based on the tracking error. We use a model-based continuous method to achieve accurate results and, in cases of failure, we re-initialize the model using our discrete tracker. This method maintains the accuracy of a more tightly coupled system, while increasing its efficiency. At any given frame, our discrete tracker uses the current and several previous frames to search into a database for the best matching solution. For improved robustness, object configuration sequences, rather than single configurations, are stored in the database. We apply our framework to the problem of 3D hand tracking from image sequences and the discrimination between fingerspelling and continuous signs in American Sign Language.