Section 2: MOT 관련 문헌에 대한 간략한 검토
Section 3: 제안된 lean tracking framework Section 4: 표준 benchmark sequences에 대해 제안된 framework의 효과 입증 Section 5: 학습 결과 요약, 향후 개선

• 목차

Abstract

• multiple object tracking
• main focus: associate objects efficiently for online and realtime applications
• detection quality = key factor influencing tracking performance
• combination : Kalman Filter + Hungarian algorithm → accuracy comparable to state-of-the-art online trackers
• simplicity → over 20x faster than other state-of-the art trackers (260Hz)

1. Introduction

MOT(Multiple Object Tracking)

• lean implementations of a tracking-by-detection framework for MOT problem
• targeted towards online tracking : only detections from the previous and the current frame are presented to the tracker
• efficiency for facilitating realtime tracking
• MOT problem can be data association problem.
• aim: associate detections across frames ni a video sequence
• tracker use various methods for modelliing the motion and appearance of obejcts in the scene

The methods of this paper were motivated through observations made on a recently established visual MOT benchmark.