MCPT: Multi-Camera People Tracking
- 주제(키워드) Object Detection , Multi Object Tracking
- 주제(DDC) 621.39
- 발행기관 아주대학교
- 지도교수 김재훈
- 발행년도 2022
- 학위수여년월 2022. 2
- 학위명 석사
- 학과 및 전공 일반대학원 컴퓨터공학과
- 실제URI http://www.dcollection.net/handler/ajou/000000031814
- 본문언어 영어
- 저작권 아주대학교 논문은 저작권에 의해 보호받습니다.
초록/요약
People tracking and detection is a fundamental problem and goal that has applications in numerous fields. The rise of machine learning methods has put Computer Vision as the forefront method to tackle this problem. But it remains a challenging problem, due to the various variables in complex real-world scenes such as the number of involved people, the possibility of multiple occlusions, and the presence of cluttered and changing backgrounds. Modern people detectors show remarkable ability in locating and detecting pedestrians even in densely populated, cluttered scenes, but they still suffer from false positives and might come with other restrictions such as having a high computational cost to run. People tracking remains a task of high interest because of its many applications to multiple different industries and fields of research such as retail, security, or automation to name a few. But for people tracking to become widely and effectively used across these fields certain conditions must be met beyond just as a theoretical or research topic. The purpose of this thesis is to explore how to enable practical people tracking for real world use. That is a solution to people tracking that delivers acceptable results as cheaply as possible. Thus, in this paper we will be thoroughly exploring the tradeoff between performance and computational cost of different people tracking models and looking for ways to reduce that cost as much as possible with minimum sacrifices in performance. Such as through using Distributed Systems and simplified tracking paradigms.
more목차
1. Introduction 1
1.1. Motivation 1
1.2. Challenges 2
1.3. Contributions 3
1.4. Structure of the Thesis 3
2. Background 4
2.1. Computer Vision 4
2.1.1. Artificial Neural Networks – ANN 4
2.1.2. Convolutional Neural Networks 8
2.1.3. Object Detection 11
2.1.4. Object Tracking 11
2.2. Modern Detection/Tracking Models 12
2.2.1. R-CNN 12
2.2.2. YOLO 14
2.2.3. Tracking Without Bells and Whistles 15
2.2.4. FEELVOS 17
3. Method 19
3.1. System Overview 19
3.2. Tracking 20
3.3. Camera Calibration 21
4. Experiments 23
4.1. Implementation 23
4.1.1. Detection 23
4.1.2. Tracking 25
4.1.3. Web Backend 25
4.2. Comparison of Tracking Methods 27
4.3. Qualitative Results 28
5. Conclusion 32
Bibliography 34
