[论文解读] Multiple Object Tracking: A Review.

Wenhan Luo, Xiaowei Zhao+1 more|arXiv (Cornell University)|Sep 26, 2014

Video Surveillance and Tracking Methods被引用 62

一句话总结

本文对计算机视觉中的多目标跟踪（MOT）进行了全面、系统的综述，从跟踪场景、方法类别、系统设计考量和评估协议等关键系统方面对方法进行了分类。它系统地阐述了MOT的原理、进展与局限性，同时指出了未来研究的开放性方向。

ABSTRACT

Multiple Object Tracking (MOT) is an important computer vision task which has gained increasing attention due to its academic and commercial potential. Although different kinds of approaches have been proposed to tackle this problem, it still has many issues unsolved. For example, factors such as continuous appearance changes and severe occlusions result in difficulties for the task. In order to help the readers understand and learn this topic, we contribute a comprehensive and systematic review. We review the recent advances in various aspects about this topic and propose some interesting directions for future research. To our best knowledge, there has not been any review about this topic in the community. The main contri-bution of this review is threefold: 1) All key aspects in the multiple object tracking system, including what scenarios the researchers are working on, how their work can be categorized, what needs to be considered when developing a MOT system and how to evaluate a MOT system, are discussed in a clear structure. This review work could not only provide researchers, especially new comers to the topic of MOT, a general understanding of the state-of-the-arts, but also help them to comprehend the aspects of a MOT system and the inter-connected aspects. 2) Instead of listing and summarizing individual publications, we categorize the approaches in the key aspects involved in a MOT system. In each aspect, the methods are divided into different groups and each group is discussed in details for the principles, advances and drawbacks. 3) We provide some potential directions with insights for MOT, which are still open issues and need more research efforts. This would be helpful for researchers to identify further interesting problems.

研究动机与目标

• 为研究人员和初学者提供多目标跟踪（MOT）的系统化、全面概述。
• 根据关键系统方面（如跟踪场景、方法组、设计考量和评估指标）对MOT方法进行分类。
• 识别并讨论MOT中尚未解决的挑战及有前景的未来研究方向。

提出的方法

• 本文系统地将MOT研究组织为关键方面：跟踪场景、方法分类、系统设计考量和评估框架。
• 根据底层原理（如外观建模、运动建模和数据关联策略）将MOT方法划分为不同类别。
• 对每一类方法进行详细分析，突出其原理、技术进展及固有局限性。
• 评估现有评估协议和基准，阐明MOT性能的衡量方式。
• 综合各类方法的见解，揭示系统组件之间的相互关联。
• 基于MOT中尚未解决的挑战，识别开放性问题并提出新的研究方向。

实验结果

研究问题

• RQ1 如何在关键系统组件上对多目标跟踪方法进行系统性分类？
• RQ2 MOT中的主要挑战（如外观变化和遮挡）是什么，当前方法如何应对？
• RQ3 构建鲁棒MOT系统时的关键设计考量有哪些？
• RQ4 MOT系统目前如何评估，现有评估协议存在哪些局限性？
• RQ5 MOT中最具前景但尚未解决的研究方向是什么？

主要发现

• 本综述是首个全面、系统化的MOT综述，填补了研究社区中的重要空白。
• 它根据核心原理对MOT方法进行了清晰分类，有助于更好地理解方法间的差异与权衡。
• 综述强调了外观变化和遮挡等重复性挑战是MOT性能的关键障碍。
• 它指出评估协议是关键但发展不足的领域，亟需标准化和优化。
• 本文概述了若干开放性研究方向，包括对长期遮挡的改进处理及对外观变化的更强鲁棒性。
• 所提出的结构化框架使研究人员能够更有效地导航MOT研究领域，并识别有意义的未来研究问题。