Nghiên cứu về kỹ thuật học sâu cho nhận diện hành động con người từ dữ liệu xương

Chuyên ngành

Computer Engineering

Người đăng

Ẩn danh

Thể loại

doctoral dissertation

2022

130
2
0

Phí lưu trữ

35 Point

Mục lục chi tiết

DECLARATION OF AUTHORSHIP

ACKNOWLEDGEMENT

ABSTRACT

CONTENTS

1. CHƯƠNG 1: AN OVERVIEW ON ACTION RECOGNITION

1.1. Data modalities for action recognition

1.2. Skeleton data collection

1.3. Data collection from motion capture systems

1.4. Data collection from RGB+D sensors

1.5. Data collection from pose estimation

1.6. Skeleton-based action recognition methods

1.6.1. Handcraft-based methods

1.6.2. Joint-based action recognition

1.6.3. Body part-based action recognition

1.6.4. Deep learning-based methods

1.6.4.1. Convolutional Neural Networks
1.6.4.2. Recurrent Neural Networks

1.7. Research on action recognition in Vietnam

1.8. Conclusion of the chapter

2. CHƯƠNG 2: JOINT SUBSET SELECTION FOR SKELETON-BASED HUMAN ACTION RECOGNITION

2.1. Preset Joint Subset Selection

2.2. Spatial-Temporal Representation

2.3. Dynamic Time Warping

2.4. Fourier Temporal Pyramid

2.5. Automatic Joint Subset Selection

2.6. Joint weight assignment. Most informative joint selection

2.7. Human action recognition based on MIJ joints

2.7.1. Preset Joint Subset Selection

2.7.2. Automatic Joint Subset Selection

2.8. Conclusion of the chapter

3. CHƯƠNG 3: FEATURE FUSION FOR THE GRAPH CONVOLUTIONAL NETWORK

3.1. Related work on Graph Convolutional Networks

3.2. Conclusion of the chapter

4. CHƯƠNG 4: THE PROPOSED LIGHTWEIGHT GRAPH CONVOLUTIONAL NETWORK

4.1. Related work on Lightweight Graph Convolutional Networks

4.2. Conclusion of the chapter

CONCLUSION AND FUTURE WORKS

ABBREVIATIONS

SYMBOLS

LIST OF TABLES

LIST OF FIGURES

Luận văn thạc sĩ a study on deep learning techniques for human action representation and recognition with skeleton data