目录

ECCV 2022

ECCV 2020

ICCV 2023

CVPR 2023

CVPR 2022

---

ECCV 2022

1.（MOT、指标）MOTCOM: The Multi-Object Tracking Dataset Complexity Metric

（包括fish datasets, sonar video, 和MOT）Caltech Fish Counting Dataset: A Benchmark for Multiple-Object Tracking and Counting

3. (multi-person MOT, dataset) Large-Scale Real-World Multi-Person Tracking

Graph-based neural network techniques

5. （SOT与MOT）Towards Grand Unification of Object Tracking 采用了相同的模型参数设置，并基于同一个网络架构实现了对多个跟踪任务的高效处理（包括SOT、MOT、VOS及MOTS）。

该算法基于（MOT）ByteTrack: Multi-Object Tracking by Associating Every Detection Box提出了一种简洁高效的数据配准方法

第7项（MOT）。[Robust Multi-Object Tracking by Marginal Inference](https://www.ecva.net/papers/eccv_2022/papers_ECCV/html/317_ECCV_2022_paper.php "Robust Multi-Object Tracking by Marginal Inference)

（MOT、3d）PolarMOT: How Far Can Geometric Relations Take Us in 3D Multi-Object Tracking? 仅限于利用几何信息进行数据匹配，并结合图神经网络方法来实现关联数据的处理

Re-examining Particle Video: Through the Analysis of Point Trajectories for Tracking through Occlusions

该文提出了一种基于Transformer架构的对象跟踪算法。论文标题为《Object Tracking via Pixel-wise Distribution Estimation》，其创新点在于将Transformer模型应用于像素级分布估计任务中，并取得了显著的跟踪性能提升效果。

（RGB-D摄像头、数据集合、三维）基于通用的RGB-D视频中的三维跟踪技术：基准与基准线

该研究报告详细阐述了基于注意力机制的AiATrack算法及其在Transformer架构下的视觉追踪应用方案

（医学）鲁棒的基于标记的支架跟踪在X射线 fluoroscopy中的应用 图卷积神经网络

该论文提出了一种基于基础组件构建的高效视觉目标跟踪架构。

基于层次化的特征嵌入机制

（MOT）Tracking Every Thing in the Wild 提出了一种新的评估标准以准确评估所有事件的表现（TETA）

17.Towards Sequence-Level Training for Visual Tracking 利用强化学习算法的视觉追踪系统采用序列级别的训练策略

第18项（分割）Robust Visual Tracking by Segmentation

该系统基于多种参数组合的混合型非局部位置与局部位置优化策略

FEAR: Lightning-fast execution, Energy-efficient, State-of-the-art accuracy and Uncompromising robustness Visual Tracking System

（平面跟踪）HVC-Net整合了Homography、Visibility和Confidence Learning以实现Planar Object Tracking，并提供了详细的论文链接：https://www.ecva.net/papers/eccv_2022/papers_ECCV/html/7396_ECCV_2022_paper.php ""HVC-Net: Unifying Homography, Visibility, and Confidence Learning for Planar Object Tracking""

22.Tracking by Associating Clips 将单一长视频序列分解为多个较短的片段，并在此基础上分别在各片段内部及相邻片段间实施目标追踪

基于 MOT 和建模的 [MOTR: End-to-End Multiple-Object Tracking with Transformer] 模型（链接）

该研究团队提出了一个基于卡尔曼滤波的平滑算法，在存在噪声干扰以及数据缺失的情况下，能够准确重构复杂且动态变化的场景中的视频图谱关键点连接关系和拓扑结构

（3D、交通场景）SpOT: Spatiotemporal Modeling for 3D Object Tracking 采用时间戳序列与边界框组合的方式表征追踪目标，并将其重新建模为空间-时间问题。该方法还利用场景参与者所具有的空间与时间信息进行建模

contributors）["BodySLAM: 同时实现相机定位、地图构建以及人体运动追踪"](https://www.ecva.net/papers/eccv_2022/papers_ECCV/html/7926_ECCV_2022_paper.php ""BodySLAM: 同时实现相机定位、地图构建以及人体运动追踪"")

人类[AvatarPoser: Articulated full-body pose tracking mechanism based on sparse motion sensing](https://www.ecva.net/papers/eccv_2022/papers_ECCV/html/74_ECCV_2022_paper.php "AvatarPoser: Articulated full-body pose tracking mechanism based on sparse motion sensing)

该系统基于三维SiameseTransformer架构设计，并结合Lidar感知技术实现目标跟踪功能

ECCV 2020

A novel approach to segment objects using point-based representations has been developed to enhance real-time multi-object tracking and segmentation processes. This innovative method focuses on improving the efficiency of online tracking systems while maintaining high accuracy in segmentation tasks. By representing segmentation through point-based methods, the system achieves faster processing times without compromising the quality of the results. The algorithm is designed to handle multiple objects simultaneously, ensuring smooth operation in dynamic environments. This advancement contributes to more effective solutions for real-world applications requiring precise object tracking and segmentation.

将物体跟踪视为几何视角下的点

3. TAO: An Extensive Evaluation Framework for Object-Oriented Tracking

第4章 针对实时性多目标跟踪

5. Object Detection Based on Spatio-Temporal Networks Aimed at Forecasting Future Positions (https://www.ecva.net/papers/eccv_2020/papers_ECCV/html/3884_ECCV_2020_paper.php "Object Tracking using Spatio-Temporal Networks for Future Prediction Location")

第6章 Adversarial-Resilient RGBT Tracking

7. PG-Net: Pixel-to-Global Matching Network for Vision-Based Tracking Process

Understanding the Environment: Leveraging Scene Data in Object Detection

Integrated Detection and Tracking of Moving Objects Using Motion Modeling Techniques for Multi-Object Tracking

10. Tracking Develops Through Examination of Static Scenes, Employing Neural 3D Mapping

ICCV 2023

Tracking by 3D Model Estimation of Unknown Objects in Videos

TrackFlow: Multi-Object Tracking with Normalizing Flows

3.An Efficient Integrated System for Three-Dimensional Object Detection and Tracking

MixCycle: 一种结合循环一致性约束的辅助半监督三维单目标跟踪方法

Foreground/Background Distribution Modelling Transformer of Visual Object Tracking

6. The paper introduces a method for tracking any object through [Decoupled Video Segmentation Technique], which effectively separates motion from appearance, enabling precise tracking across various scenarios and objects in real-time video streams.

Diverse Data Representation-Driven Active Learning for Multi-Object Tracking

Goel, et al., "Four-Dimensional Human Reconstruction and Tracking Using Transformer Models," ICCV 2023

Investigating Motion-Aware Matching of Monocular 3D Object Tracking

（物体建模）该研究提出了一种鲁棒的物体建模方法用于视觉追踪

Integrating Boxes and Masks: A Multi-Object Approach for Integrated Visual Tracking and Segmentation

Centric Object-based Multiple Object Tracking

该模型基于预测轨迹假设实现了三维物体跟踪Transformer算法

Track everything everywhere at once

（该数据集）！SportsMOT: A Comprehensive Dataset Encompassing Multi-Sport Scenarios and Advanced Multi-Object Tracking Technology

17. 基于不确定性感知的无监督多目标追踪系统（Liu, 2023）

18.（无人机）[Adaptability and Background-Awareness Enhanced Vision Transformer System for Real-Time Unmanned Aerial Vehicle Tracking](https://openaccess.thecvf.com/content/ICCV2023/html/Li_Adaptive_and_Background-Aware_Vision_Transformer_for_Real-Time_UAV_Tracking_ICCV_2023_paper.html "Adaptive and Background-Aware Vision Transformer for Real-Time UAV Tracking)

19. Multi-Plane Object Tracking)

Data Set 20 is a novel reference dataset named 360VOT, designed for panoramic object tracking in visual space.

21. Tracking without supervision: Unsupervised multi-object tracking based on contrastive similarity learning
22. TAPIR: A method for tracking any point, utilizing per-frame initialization and temporal refinement

（人）TEMPO: Efficient Multi-View pose estimation, tracking and forecasting

Synchronizing Feature Extraction and Matching: A Single-Branch Framework for 3D Object Tracking (ICCV 2023)

Collaborative Learning in Tracking for Frame-Rate-Insensitive Multi-Object Tracking

（数据集）PlanarTrack: A Large-scale Challenging Benchmark for Planar Object Tracking.

Deep Active Contours for Real-Time 6 Degrees of Freedom Object Tracking

联合文本

本研究涉及的是模糊图像中人体姿态跟踪技术的应用

CVPR 2023

（Drone）[Efficiency-Oriented RGB-D Aerial Tracking](https://openaccess.thecvf.com/content/CVPR2023/html/Yang Resource-Efficient RGB-D Aerial Tracking CVPR 2023 Paper.html "Resource-Efficient RGB-D Aerial Tracking")

Representation Learning for Visual Object Tracking via Masked Appearance Transfer

TrackS: S2S Learning for Visual OT, CVPR 2023 Paper

4.Frame-Event Alignment and Fusion Network for High Frame Rate Tracking 融合传统帧和基于事件相机

5.OVTrack: Open-Vocabulary Multiple Object Tracking 在预训练外类别上的鲁棒性能以及在大规模现实环境下的能力表现

6. MotionTrack系统旨在通过学习可靠的短期和长期运动模式实现多目标跟踪功能

7.[Bridging Search Area Interactions In the Template of RGB-T Tracking](https://openaccess.thecvf.com/content/CVPR2023/html/Hui_Bridging_Search_Region_Interaction_With Template_for_RGB-T_Tracking_CVPR_2023_paper.html "Bridging Search Region Interaction With Template for RGB-T Tracking")

（密集遮挡）MotionTrack: Acquisition of Robust Short-term and Long-term Motions to Track Multiple Objects

9.采用LIDAR技术的VoxelNeXt模型用于实现三维物体检测与追踪

multi-cameras系统处于过去与未来之间：一种基于空间-时间建模的多摄像头三维多物体跟踪方法

Emphasize Details: Online Multi-Object Tracking With a variety of fine-grained representations.

（细胞）[无监督的边缘跟踪方法用于活细胞基于机械力和周期一致性损失]

(13)GarmentTracking: Category-Level Garment Pose Tracking

VideoTrack: Object tracking methods have achieved remarkable success through the use of video transformers.

15.（基于LiDAR技术和相关数据集）[ARKitTrack: A New Diverse Dataset for Tracking Using Mobile RGB-D Data](https://openaccess.thecvf.com/content/CVPR2023/html/Zhao_ARKitTrack_A_New_Diverse_Dataset_for_Tracking_Using_Mobile_RGB-D_CVPR_2023_paper.html "ARKitTrack: A New Diverse Dataset for Tracking Using Mobile RGB-D Data)

UTM: 一个整合多目标跟踪模型通过身份意识的特征增强

17. # Integrating Short- and Long-Term Object Tracking via Graph-Based Hierarchy](https://openaccess.thecvf.com/content/CVPR2023/html/Cetintas_Unifying_Short_and_Long-Term_Tracking_With_Graph_Hierarchies_CVPR_2023_paper.html "Unifying Short and Long-Term Tracking With Graph Hierarchies") 长视频物体跟踪

18. (Interaction Tracking and Reconstruction) A Novel Framework for Visibility-Aware Human-Object Interaction Tracking from a Single RGB Camera)

19.（多模态文本定位）基于自然语言描述的联合视觉定位与追踪系统

Observation-Centric SORT: Rethinking SORT for Robust Multi-Object Tracking 优化基于观测的SORT算法以提升多目标跟踪的鲁棒性

该系统实现了多模态跟踪技术

该方法采用基于神经网络的六自由度追踪系统来实现三维重建未知物体的过程。

23.(数据集、动物轨迹跟踪）3D-POP - An Automated Annotation Approach to Facilitate Markerless 2D-3D Tracking of Freely Moving Birds With Marker-Based Motion Capture 用运动捕捉（mo-cap）系统，以半自动方式获取大量动物运动和姿势（二维和三维）注释数据

事件摄像机跟踪器

（数据集）JRDB-Pose: A Comprehensive Database for Extensive Multi-Person Pose Estimation and Tracking

（集成多模态文本处理系统）Referring Multi-Object Tracking

Autoregressive-based visual tracking technology)

（虚拟版本）Avatars Grow Legs: Generating Smooth Human Motion From Sparse Tracking Inputs With Diffusion Model

MOTRv2: Bootstrapped End-to-End Multi-Object Tracking Using Pretrained Object Detectors

CVPR 2022

该系统通过整合区域与深度信息实现了对无纹理物体的高效三维追踪。该系统通过整合区域与深度信息实现了对无纹理物体的高效三维追踪

（开放世界）OpenWorld Tracking系统及其应用前景进行了详细介绍，并探讨了如何在具有开放世界的环境下进行检测与跟踪。

3.（基于无人机的多目标跟踪系统）该论文探讨了将多目标跟踪技术与动态无人机平台相结合的方法

4.（人机互动、数据集）[BEHAVE: Data Set and Approach to Monitor Human-Object Interaction](https://openaccess.thecvf.com/content/CVPR2022/html/Bhatnagar_BEHAVEMONITOR_Dataset_and_method_for_tracking_human_object_interactions_CVPR_2022_paper.html "BEHAVEMONITOR: Data Set and Approach to Monitor Human-Object Interaction)

5.（人）Tracking People by Predicting 3D Appearance, Location and Pose

Adapting Model Output in the context of Tracking

MeMOT: Multi-Object Tracking With Memory

8. (动物行为分析）在动物行为分析领域，《PyMiceTracking: An Open-Source Toolbox for Real-Time Behavioral Neuroscience Experiments》这一开源工具包提供了实时的行为神经科学研究实验解决方案，并可访问于《CVPR 2022》会议论文集中的相关页面

该系统采用先进的追踪机制结合高效的目标分割算法进行分析，并通过智能匹配策略实现精确的实例识别功能

小轨迹查询，端到端的视频实例分类、分割、跟踪

Global Tracking Transformers

基于全局多目标追踪的方案

该基准集合包含基于真实3D的数据集、SOT方法以及 markerless 的高精度三维物体跟踪基准集

该文提出了一种基于无监督学习的准确的Siamese跟踪方法

13.（SOT与MOT均可）Unified Transformer Tracker for Object Tracking

该系统能够实时提取并整合时域与空间域的信息

（夜晚空域追踪与数据集）Unsupervised Domain Adaptation for Nighttime Aerial Tracking

（UAV）TCTrack: Temporal Information for Aerial Object Tracking 通过时间信息来提升空间特征

（自动驾驶、轨迹预测）Is it whoever's track anyway? By enhancing the robustness to tracking errors using affinity-based trajectory forecasting, this work achieves more reliable results in motion prediction tasks.

（自动驾驶、单目视频）[Time3D: Simultaneous 3D Object Detection and Tracking for Autonomous Driving](https://openaccess.thecvf.com/content/CVPR2022/html/Li_Time3D_End-to-End_Joint_Monocular_3D_Object_Detection_and_Tracking_for_CVPR_2022_paper.html "Time3D: End-to-End Joint Monocular 3D Object Detection and Tracking for Autonomous Driving)

（MOT） Adiabatic Quantum Computing Framework focused on Multiple Object Tracking, presented at the CVPR 2022 conference, available at https://openaccess.thecvf.com/content/CVPR2022/html/Zaech_Adiabatic_Quantum_Computing_for_Multi_Object_Tracking_CVPR_2022_paper.html "Adiabatic Quantum Computing for Multi Object Tracking")

（在线、MOT）Towards Discriminative Representation: A method that employs multi-view trajectory contrastive learning to achieve online multi-object tracking 多视角轨迹对比学习

该系统采用组合提升分割算法与几何投影方法相结合的方式实现了高效的目标跟踪效果

22.（手部、视频数据集）

卷积结构实现大规模非受限视频中手的位置以及运动轨迹的定位和追踪

23.Correlation-Aware Deep Tracking

基于Transformer的跟踪技术结合创新性架构的多尺度循环移窗注意力机制

（持续监测）该论文题为《基于局部追踪器集合的全局追踪方法》Global Tracking via Ensemble of Local Trackers ，该研究综合运用了两种长期追踪策略，并通过多个追踪器协同追踪目标物体的运动轨迹

混合注意力机制（MAM），用于同时提取特征信息并整合目标相关信息

LiDAR-based PTTR: Relation-based 3D Point Cloud Object Tracking via the Transformer architecture

（MOT、3d）TrackFormer: Multi-Object Tracking With Transformers该系统能够避免任何额外的图形优化工作以及无需考虑任何额外的运动相关的建模以及外观相关的建模

该文系统性地构建了一个基于排序机制的优化框架，并在公开数据集上进行了广泛的实验验证

开发出一种卓越的双阶段追踪系统，在LIDAR与三维结合的技术下，在点云中实现了对三维单目标追踪的高度准确。（lidar、3d）Beyond 3D Siamese Tracking: A Motion-Centric Paradigm for 3D Single Object Tracking in Point Clouds

基于最小成本流（MCF）建模的multi-object tracking问题

（人员、MOT、数据库）[PoseTrack ̈̈̈̈̈̈̈̈̈̈ 1: 一个人脸检索、多目标跟踪及多人姿态跟踪的数据集](https://openaccess.thecvf.com/content/CVPR ̈̈̈̈̈̈̈̈ ² ₀ ₂ ₂/ html /Doring_PoseTrack ₁ ² ₁:A_Dataset_for(PersonaltySearch,Multi-Objective_Tracking_and_Multi-People_Pose_Tracking)CVPR ₂ ₀ ₂ ₂_paper.html "Personalty检索、多目标跟踪及多人姿态跟踪的数据集)

（包括可见光温度RGBT技术、无人机（UAV）以及相关数据集）Visible Thermal UAV Tracking: A Comprehensive Benchmark and Baseline System

该数据集基于"multi-person tracking"框架，并结合"MOT"技术进行设计与实现；其核心创新点在于通过统一外观特征与多样的运动模式相结合的方式进行目标追踪。