[toc]
Comments: Accepted Manuscript in IEEE Transactions on Intelligent Transportation Systems
-
[Submitted on 2 Dec 2019 (v1), last revised 26 Jan 2021 (this version, v2)]
Abstract
- benchmark dataset
- evaluation metrics
- fundamental characteristics, primary motivations, and contributions of DL-based methods
- conducting critical analyses quantitatively and qualitatively, their pros and cons under various common scenarios are investigate
introduction
- 任务：只给出目标的初始状态，来估计一个视觉目标的未知的轨迹。
- 应用广泛：自动驾驶、自动机器人、监控、增强现实、无人机跟踪、运动、手术、生物、海洋探险。
- ill-posed defifinition of the visual tracking((i.e., model-free tracking, on-the-flfly learning, single-camera, 2D information)
- 在现实世界中的challenge（ay include arbitrary classes of targets (e.g., human, drone, animal, vehicle) and motion models, various imaging characteristics (e.g.,static/moving camera, smooth/fast movement, camera resolution), and changes in environmental conditions (e.g., illumination variation, background clutter, crowded scenes). 包括任意类别的目标(如人类、无人机、动物、车辆)和任意类别的运动模型，各种成像特征（静态/移动相机，平稳/快速运动，相机的分辨率)，以及环境条件的变化(例如，光照变化，背景杂乱，拥挤场景)。