Call number:
S 99.0139(362)
In:
Wissenschaftliche Arbeiten der Fachrichtung Geodäsie und Geoinformatik der Leibniz Universität Hannover, Nr. 362
Type of Medium:
Series available for loan
Pages:
XV, 143 Seiten
,
Illustrationen, Diagramme
Series Statement:
Wissenschaftliche Arbeiten der Fachrichtung Geodäsie und Geoinformatik der Leibniz Universität Hannover Nr. 362
Language:
English
Note:
1 Introduction
1.1 Contributions
1.2 Thesis outline
2 Basics
2.1 Convolutional Neural Networks
2.1.1 Training
2.1.2 CNN Architectures
2.2 Active Shape Model
2.3 Monte Carlo based optimisation
3 State of the art
3.1 Data driven approaches
3.1.1 Viewpoint prediction
3.1.2 3D pose prediction
3.1.3 3D pose and shape prediction
3.2 Model driven approaches
3.2.1 Shape priors
3.2.2 Scene priors
3.2.3 Shape aware reconstruction
3.2.4 Optimisation
3.3 Discussion
4 Methodology
4.1 Overview
4.1.1 Input
4.1.2 Problem statement
4.1.3 Scene layout
4.1.4 Detection of vehicles
4.2 Subcategory-aware 3D shape prior
4.2.1 Geometrical representation
4.2.2 Mode Learning
4.3 Multi-Task CNN
4.3.1 Input branch
4.3.2 Vehicle type branch
4.3.3 Viewpoint branch
4.3.4 Keypoint/Wireframe branch
4.3.5 Training
4.4 Probabilistic vehicle reconstruction
4.4.1 3D likelihood
4.4.2 Keypoint likelihood
4.4.3 Wireframe likelihood
4.4.4 Position prior
4.4.5 Orientation prior
4.4.6 Shape prior
4.4.7 Inference
4.5 Discussion
5 Experimental setup
5.1 Objectives
5.2 Test data
5.2.1 KITTI benchmark
5.2.2 ICSENS data set
5.3 Parameter settings and training
5.3.1 Learning the ASM
5.3.2 Training of the CNN
5.4 Evaluation strategy and evaluation criteria
5.4.1 Detection
5.4.2 Multi-Task CNN
5.4.3 Probabilistic model for vehicle reconstruction
5.4.4 Comparison to related methods
6 Results and discussion
6.1 Detection
6.2 Evaluation of the CNN components
6.2.1 Evaluation of the viewpoint branch
6.2.2 Evaluation of the vehicle type branch
6.3 Ablation studies of the model components
6.3.1 Analysis of the observation likelihoods
6.3.2 Analysis of the state priors
6.4 Analysis of the full model for vehicle reconstruction
6.4.1 Evaluation of the pose
6.4.2 Evaluation of the shape
6.4.3 Analysis of further aspects
6.5 Comparison to related methods
6.6 Discussion
6.6.1 Likelihood terms
6.6.2 State priors
6.6.3 Full model
6.6.4 Inference
7 Conclusion and outlook
,
Sprache der Kurzfassungen: Englisch, Deutsch
Location:
Lower compact magazine
Branch Library:
GFZ Library
Permalink