Frontiers of Computer Vision: 26th International Workshop, IW-FCV 2020, Ibusuki, Kagoshima, Japan, February 20–22, 2020, Revised Selected Papers ... in Computer and Information Science, 1212)

Preface
Organizing Committee
Contents
Real-World Applications
Efficient and Fast Traffic Congestion Classification Based on Video Dynamics and Deep Residual Network
	1 Introduction
	2 The Proposed System
		2.1 Video Dynamics Extraction
		2.2 Feature Extraction Using Deep CNN
		2.3 Classification Step
	3 Experimental Results and Discussion
		3.1 Analyzing the Performance of the Proposed System with UCSD Dataset
		3.2 Analyzing the Performance of the Proposed System with NU1 Dataset
		3.3 Analyzing the Processing Time of the Proposed System
	4 Conclusion
	References
Early Wildfire Detection Using Convolutional Neural Network
	1 Introduction
	2 Dataset Collection
		2.1 Automatic Image Crawling
		2.2 Automatic and Manual Cleanup
		2.3 Final Patch and Class Labeling
	3 Proposed Framework
		3.1 Network Architecture Selection
		3.2 Class Imbalance
		3.3 Training
		3.4 Implementation Details
	4 Experimental Results and Discussion
		4.1 Multi-class Classification
		4.2 Binary Projection
		4.3 Optimal Cut-Off Selection
		4.4 Patch Classification to Frame-Level Detection
		4.5 Qualitative Wildfire Detection Results on Unseen Data
	5 Conclusions
	References
Deep Matting for AR Based Interior Design
	1 Introduction
	2 Related Work
	3 Proposed Methodology
		3.1 ROI Selection
		3.2 Handling Illumination Changes
		3.3 Deep Foreground/Background Objects Segmentation
		3.4 Deep Alpha-Matting
		3.5 User Selected Texture Transformation
	4 Experiments
		4.1 Deep Image Matting Evaluation
		4.2 Deep Foreground Segmentation Evaluation
		4.3 User\'s Qualitative IID Experience
	5 Conclusion
	References
Examination and Issues of Kumamoto Castle Ishigaki Region Extraction Focusing on Stone Contour Features
	1 Introduction
	2 Related Research
	3 Stone Extraction Method
		3.1 GrabCut Extraction
		3.2 Watershed Extraction
		3.3 GrabCut and Watershed Extraction
	4 Experiment
		4.1 Experiment Environment
		4.2 Error Calculation with Ground Truth
		4.3 Extraction Results Obtained by Each Method
		4.4 Selecting the Contour Candidate with Minimum Error
		4.5 Consideration
	5 Conclusion
	References
Detection of Speech Impairments in Parkinson Disease Using Handcrafted Feature-Based Model on Spanish Speech Corpus
	1 Introduction
	2 Related Work
	3 Proposed Methodology
		3.1 Handcrafted Feature Extraction
		3.2 Classification
	4 Experimental Setup and Results
		4.1 Experimental Tools
		4.2 Spanish Speech Dataset
		4.3 Evaluation Metrics
		4.4 Results
	5 Conclusion
	References
Face, Pose, and Action Recognition
Short-Term Action Recognition by 3D Convolutional Neural Network with Pixel-Wise Evidences
	1 Introduction
	2 Related Works
		2.1 3D Convolutional Neural Network
		2.2 Autoencoder
		2.3 C3D
		2.4 GoogLeNet
	3 3D CNN with Pixel-Wise Evidences
	4 Experiments
		4.1 Dataset
		4.2 Implementation Details of PWE 3D CNN
		4.3 Performance of the PWE 3D CNN
		4.4 Comparisons with C3D Model
		4.5 Comparison with the State-of-the-art
		4.6 Ablation Experiments
	5 Conclusions
	References
Discriminative Metric Learning with Convolutional Feature Descriptors for Age-Invariant Face Recognition and Verification
	1 Introduction
	2 Previous Work
	3 DML with Convolutional Feature Descriptors
		3.1 Preprocessing
		3.2 Convolutional Feature Descriptors
		3.3 Discriminative Metric Learning
	4 Evaluation Experiments
		4.1 Datasets
		4.2 Network Implementation
		4.3 Results and Discussion
	5 Conclusions
	References
Dilated CNN Based Human Verifier for Intrusion Detection
	1 Introduction
	2 Proposed Algorithm
		2.1 Probabilistic Change Detector (PCD)
		2.2 Dilated CNN
	3 Experimental Analysis
		3.1 Dataset Description
		3.2 Parameter Setting
		3.3 Quantitative Analysis
		3.4 Qualitative Analysis
	4 Conclusion
	References
Occlusion-Aware Skeleton Trajectory Representation for Abnormal Behavior Detection
	1 Introduction
	2 Related Work
	3 Abnormal Behavior Detection Making Use of Skeleton Representations
		3.1 Occlusion-Aware Skeleton Trajectory Representation
		3.2 Implementation Details
	4 Experiment
		4.1 Drunk Walking Dataset
		4.2 Comparison Methods
		4.3 Network Parameters
		4.4 Experimental Setting
		4.5 Results
		4.6 Performance on Data with Different Levels of Missing Joints
	5 Conclusion
	References
A Deep-Learning Based Worker’s Pose Estimation
	1 Introduction
	2 Related Work
	3 Worker Pose Estimation System
		3.1 Architecture of the Proposed System
	4 Neck and Wrist Angle Calculations
	5 Experimental Results
		5.1 Discussion and Analysis
	6 Limitations and Ongoing Work on the Proposed System
	7 Conclusion
	References
Identifying People Using Body Sway in Case of Self-occlusion
	1 Introduction
	2 The Influence of Self-occlusion
	3 Our Method
		3.1 Overview
		3.2 Estimating Head Regions from a Set of Images of a Person
		3.3 Extracting a Spatio-Temporal Feature from Silhouette Images of Head Regions
	4 Experiments
		4.1 Dataset
		4.2 Assessing Accuracy of Estimating Head Regions
		4.3 Evaluation of Identification Performance
		4.4 Performance Comparison When Using Spatial Features and Temporal Features
		4.5 Comparison of Proposed Method with Prevalent Methods
	5 Conclusions
	References
Action Recognition in Sports Video Considering Location Information
	1 Introduction
	2 Previous Work
	3 Proposed Method
		3.1 Shooting Method
		3.2 Conversion to Overhead Image
		3.3 Handcraft Features
		3.4 Heatmap Features
		3.5 Use of Heatmap Features and Subdivision Areas
	4 Experiment
		4.1 Dataset Details
		4.2 Evaluation Index
	5 Discussion
		5.1 Handcraft Features and Heatmap Features
		5.2 Subdivision of Play Area
	6 Conclusion
	References
Object Detection and Tracking
Adaptive Feature Selection Siamese Networks for Visual Tracking
	1 Introduction
	2 Related Work
		2.1 Siamese Based Trackers
		2.2 Attention-Based Trackers
	3 Proposed Tracking Framework
		3.1 Fully Convolutional Siamese Network
		3.2 Adaptive Feature Selection Siamese Tracker
		3.3 Network Training
	4 Experiments
		4.1 Comparison with State-of-the-Art Trackers
		4.2 Experiments on OTB50 and OTB100
		4.3 Experiments on TC128
		4.4 Experiments on VOT2017
	5 Conclusion
	References
Faster R-CNN with Attention Feature Map for Robust Object Detection
	1 Introduction
	2 Related Works
	3 Proposed Method
		3.1 Extraction for Attention Feature Map
		3.2 Multi Task Loss
		3.3 Training
	4 Experiment
		4.1 Experimental Setup
	5 Conclusion
	References
Indoor Visual Re-localization Based on Confidence Score Using Omni-Directional Camera
	1 Introduction
	2 Related Work
	3 Proposed Method
		3.1 System Overview
		3.2 PoseNet and Bayesian PoseNet
		3.3 Data Augmentation with Semi-Omni-Directional Image
		3.4 Localization Method
	4 Experiment
		4.1 Dataset
		4.2 Experimental Setting
		4.3 Experimental Result
	5 Discussion
	6 Conclusion
	References
Analysis of Information Flow in Hidden Layers of the Trained Neural Network by Canonical Correlation Analysis
	1 Introduction
	2 Method
		2.1 Canonical Correlation Analysis
		2.2 CCA for Information Flow Analysis
	3 Experiments and Results
		3.1 White Balancing by CNN
		3.2 Multi-task Learning
	4 Conclusion
	References
Inspection and Diagnosis
Study of GANs Using a Few Images for Sealer Inspection Systems
	1 Introduction
	2 Methods
		2.1 Generative Adversarial Networks
		2.2 Evaluation
	3 Results and Discussion
	4 Conclusion
	References
Consistency Ensured Bi-directional GAN for Anomaly Detection
	1 Introduction
	2 Related Work
	3 Proposed Method
		3.1 Model Overview
		3.2 Consistency of Mappings Between Image Space and Latent Space
		3.3 Conditioning of Image with Latent Variables in Bi-GAN Architecture
		3.4 Model Training
		3.5 Anomaly Score
	4 Experiments
		4.1 MNIST Dataset
		4.2 MVTec Anomaly Detection Dataset
	5 Conclusion
	References
Unsupervised Adversarial Learning for Dynamic Background Modeling
	1 Introduction
	2 Related Work
	3 Proposed Methodology
		3.1 BI-GAN Training
		3.2 BI-GAN Testing
	4 Implementation of BI-GAN
	5 Experiments
		5.1 Evaluation of BI-GAN on SBM.net Dataset
		5.2 Evaluation of BI-GAN on SBI Dataset
		5.3 Failure Cases
	6 Conclusion
	References
Transfer Learning by Cascaded Network to Identify and Classify Lung Nodules for Cancer Detection
	1 Introduction
	2 Materials and Methods
	3 Cascaded Architecture
		3.1 Segmentation Network
		3.2 Classification Network
	4 Experimental Settings
		4.1 Experiment
		4.2 Evaluation
	5 Results
	6 Conclusion
	References
Hybrid Deep Learning and Data Augmentation for Disease Candidate Extraction
	1 Introduction
	2 Proposed Segmentation Architecture
	3 Data Augmentation for Deep Learning
	4 Evaluation
	5 Conclusion
	References
Camera, 3D and Imaging
Multispectral Photometric Stereo Using Intrinsic Image Decomposition
	1 Introduction
	2 Related Work
	3 Image Formulation
	4 Multispectral Color Photometric Stereo Method
		4.1 Channel Crosstalk
		4.2 Edge of Multiple Albedos
		4.3 Intrinsic Image Decomposition
		4.4 Photometric Linearization
		4.5 Surface Normal Estimation
		4.6 Calculating Height from Surface Normal
	5 Experiment
		5.1 Experimental Setup
		5.2 Experimental Result
	6 Conclusion
	References
In-Plane Rotation-Aware Monocular Depth Estimation Using SLAM
	1 Introduction
	2 Related Work
		2.1 In-Plane Rotation-Aware Prediction
		2.2 SLAM with Monocular Depth Estimation
		2.3 Learning Based Rotation Prediction
	3 Method
		3.1 Camera Pose Estimation
		3.2 CNN Depth Estimation
		3.3 Roll Alignment
	4 Experiment
		4.1 Experiment Detail
		4.2 Dataset
	5 Result
		5.1 Qualitative Evaluation
		5.2 Quantitative Evaluation
	6 Conclusion
	References
Uncalibrated Photometric Stereo Using Quadric Surfaces with Two Cameras
	1 Introduction
	2 Formulation
		2.1 Inverse Rendering
		2.2 Quadric Surfaces and Coordinate Systems
	3 Image Generation Model
		3.1 Reflection Model
		3.2 Contour of Quadric Surface
		3.3 Normal Vector of Quadric Surface
		3.4 Projection Model
	4 Minimization Method
		4.1 Levenberg-Marquardt Method
		4.2 Elements of p
		4.3 Finite Difference for a Jacobian
		4.4 Approximate Reflectance with Average Image
		4.5 Contour Mismatch
		4.6 Diagonal Terms
	5 Experimental Results
		5.1 Datasets
		5.2 Initial Values
		5.3 Results
	6 Conclusions
	References
Gaussian Processes for Efficient Plane-Based Camera Calibration
	1 Introduction
		1.1 Our Contribution
	2 Gaussian Processes for Solving Sensor Location Problems
		2.1 Gaussian Process Model
		2.2 Gaussian Processes for Sensor Location Problems
	3 Proposed Method
		3.1 Image Feature
		3.2 Kernel Function
	4 Experimental Results
		4.1 Experiments with Synthetic Data
		4.2 Experiments with Real Data
	5 Conclusion
	Appendix A Plane-Based Camera Calibration
		Appendix A.1 Pin-Hole Camera Model
		Appendix A.2 Plane-Based Camera Calibration
	References
Leveraging Pyramidal Feature Hierarchy for 3D Reconstruction
	1 Introduction
	2 Related Work
		2.1 Single-View and Multi-view 3D Reconstruction Networks
		2.2 Pyramidal-Based Networks
	3 Approach
		3.1 Network Architecture
	4 Experiments
		4.1 Qualitative Evaluation
		4.2 Quantitative Evaluation
		4.3 Conclusion
	References
Inverse Lighting from Cast Shadows Under Unknown Radiometric Response Function
	1 Introduction
	2 Proposed Method
		2.1 Illumination Distribution
		2.2 Radiometric Response Function
		2.3 Joint Estimation
		2.4 Joint Estimation with Approximate Geometry
	3 Experiments
		3.1 Comparison Using Synthetic Images
		3.2 Comparison Using Real Images
		3.3 Sensitivity Analysis
	4 Conclusion
	References
Author Index