声明
摘要
ABSTRACT
CONTENTS
LIST OF FIGURES
LIST OF TABLES
Chapter 1 Introduction
1.1 Human attributes analysis for behaviors
1.1.1 Object based approaches
1.1.2 Holistic approaches
1.1.3 Hybrid approaches
1.2 Thesis motivation
1.3 Objectives
1.4 Innovation
1.5 Organization of the thesis
1.6 Chapter summary
1.7 本章小结
Chapter 2 Algorithm for Pedestrians’ Full Body Orientation and Direction of Attention Via Deep Learning Approach
2.1 Chapter introduction
2.2 Existing works regarding pedestrians’ full body orientation and direction of attention
2.3 Convolutional neural network(CNN)
2.3.1 Convolution layer
2.3.2 Pooling layer
2.3.3 Rectified Linear Units(ReLUs)layer
2.3.4 Fully connected layer
2.3.5 Dropout layer
2.3.6 Softmax layer
2.3.7 Training of a CNN
2.4 Proposed CNN design
2.4.1 Training datasets preparation
2.5 Results and discussions
2.5.1 Fine tuning
2.5.2 Testing protocol
2.5.3 Experiment on TUD-Multiview pedestrian dataset
2.5.4 Experiment on CAVIAR dataset
2.5.5 Experiment on real-time video sequences
2.6 Chapter summary
2.7 本章小结
Chapter 3 Algorithm for Pedestrians’Classification by UsingStacked Sparse Autoencoder
3.1 Chapter introduction
3.2 Existing works regarding pedestrian classifications
3.3 The proposed approach
3.3.1 Saliency maps
3.3.2 Sparse autoencoder(SAE)
3.3.3 SSAE
3.3.4 Softmax classifier
3.4 Results and discussions
3.5 Chapter summary
3.6 本章小结
Chapter 4 Algorithms for Distance and Dimensions Estimations of Pedestrians in Real-Time Environments
4.2 Existing works regarding distance and dimensions estimations of pedestrians
4.3 The proposed methodology
4.3.1 Pedestrian identification
4.3.2 Foreground objects extraction
4.3.3 CNN model for pedestrian identification
4.3.4 Distanee and dimensions estimations
4.4 Resuits and discussions
4.4.1 Pedestrian identification
4.4.2 Distance and dimension estimations
4.5 Chapter summary
4.6 本章小结
Chapter 5 Algorithm for Pedestrian Gender Recognition Using Stacked Sparse Autoencoders
5.1 Chapter introduction
5.2 Existing works regarding pedestrian gender recognition
5.3 Pedestrian gender recognition via stacked sparse autoencoder
5.3.1 Parsing a pedestrian
5.3.2 Pedestrian gender classification(AE)
5.4 Results and discussions
5.4.1 Training dataset
5.4.2 Training and fine-tuning of proposed SSAE
5.4.3 Testing with MIT dataset
5.4.4 Testing with PETA dataset
5.5 Chapter summary
5.6 本章小结
Chapter 6 Algorithm for Pedestrian Gender Recognition Using Convolutional Neural Networks
6.1 Chapter introduction
6.2 Pedestrian gender recognition via deep convolutional neural networks
6.2.1 Parsing pedestrians
6.2.2 Convolutional neural network(CNN)
6.3 Results and discussions
6.4 Chapter summary
6.5 本章小结
Chapter 7 Conclusions and Future Research
7.1 Summary of contributions
7.2 Future research
第7章 总结与展望
7.1 研究总结
7.2 工作展望
REFERENCES
LIST OF PUBLICATIONS
ACKNOWLEDGEMENTS