All Weather Robot Vision

Description

Complex illumination and meteorological conditions can significantly limit the robustness of robotic vision systems. This book focuses on image pre-processing for robot vision in complex illumination and dynamic weather conditions. It systematically covers cutting-edge models and algorithms, approaching them from a novel viewpoint based on studying the atmospheric physics and imaging mechanism. It provides valuable insights and practical methods such as illumination calculations, scattering modeling, shadow/highlight detection and removal, intrinsic image derivation, and rain/snow/fog removal technologies that will enable robots to be effective in diverse lighting and weather conditions, i.e., ensure their all-weather operating capacity. As such, the book offers a valuable resource for researchers, graduate students and engineers in the fields of robot engineering and computer science.     Preface Contents 1 Introduction 1.1 Robot Vision Overview 1.2 Robot Vision in Complex Illumination and Bad Weather 1.3 Color Matching Functions and Scene Reproduction 1.4 Overview of Key Technologies for All-Weather Robot Vision 1.4.1 Imaging Elements 1.4.2 Image Processing of Complex Illumination 1.4.3 Image Processing of Bad Weather 1.5 The Main Contents and Structure of This Book References 2 Spectral Power Distributions and Reflectance Calculations for Robot Vision 2.1 Introduction 2.2 Outdoor Spectral Power Distributions Calculations for Robot Vision 2.2.1 Absorption of Extraterrestrial Irradiance Passing Through the Atmosphere 2.2.2 Computing Direct Sunlight 2.2.3 Computing Diffuse Skylight 2.2.4 Experiments and Comparisons 2.3 Wavelength-Sensitive Function Controlled Reflectance Reconstruction Appendix References 3 Imaging Modeling and Camera Sensitivity Recovery 3.1 Overview of Digital Imaging 3.2 Simulation of Digital Imaging Process 3.2.1 Color Image Formation 3.2.2 Imaging Simulation 3.2.3 Camera Radiometric Calibration 3.2.4 Simulation of In-Camera Post-processing 3.3 Camera Spectral Sensitivity Recovery References 4 Shadow Modeling and Detection 4.1 Introduction 4.2 Tricolor Attenuation Model of Shadows 4.3 Tricolor Linear Model of Shadows 4.4 Shadow Detection Based on Tricolor Attenuation Model 4.5 Shadow Detection Based on Tricolor Linear Model 4.5.1 New Shadow Properties 4.5.2 Shadow Detection Algorithm 4.5.3 Experiments 4.6 Evaluation of Shadow Features 4.6.1 Features Selected for Evaluation 4.6.2 Feature Evaluation References 5 Intrinsic-Image Deriving and Decomposition 5.1 Introduction 5.2 Intrinsic-Image Deriving 5.2.1 Intrinsic-Image Extraction Based on Tricolor Linear Model 5.2.2 Intrinsic-Image Extraction Based on Tricolor Attenuation Model 5.2.3 Pixel-Wise Orthogonal Decomposition 5.3 A New Intrinsic-Lighting Color Space for Daytime Outdoor Images 5.3.1 New Color Space: IL Space 5.3.2 Shadow/Lighting Processing via the IL Space 5.3.3 Experiments for Shadow-Free Color Images 5.3.4 Intrinsic-Lighting Contour Surface Embedding and Relighting References 6 Shadow and Highlight Removal 6.1 Introduction 6.2 Shadow Removal Based on Tricolor Linear Model 6.3 Deep Learning-Based Shadow Removal 6.3.1 A New Dataset for Shadow Removal—SRD 6.3.2 Proposed Model 6.3.3 Experiments 6.4 Highlight Removal with Color-Lines Constraint 6.4.1 Global Color-Lines Constraint and Pixel Clusters 6.4.2 Separation of Specular Reflection 6.4.3 Experiments References 7 Rain and Snow Removal 7.1 Introduction 7.2 Snowflake Removal for Videos via Global and Local Low-Rank Decomposition 7.2.1 Problem Model 7.2.2 Algorithm 7.2.3 Experiments 7.3 Video Desnowing and Deraining Based on Matrix Decomposition 7.3.1 General Model 7.3.2 Moving Objects and Sparse Rain Streaks Modeling with MRFs 7.3.3 Group Sparsity for Moving Objects 7.3.4 Algorithm 7.3.5 Experiments 7.4 Dually Connected Single-Image Deraining Net Using Pixel-Wise Attention 7.4.1 Dually Connected Deraining Net 7.4.2 Experiments Appendix References 8 Single-Image Dehazing 8.1 Introduction 8.2 Single-Image Dehazing by Latent Region Segmentation 8.2.1 Estimating Transmission 8.2.2 Optimizing Transmission 8.2.3 Recovering the Scene Radiance 8.2.4 Experimental Results 8.3 Deep Retinex Network for Single-Image Dehazing 8.3.1 Retinex-Based Dehazing Model 8.3.2 Network Architecture 8.3.3 Network Optimization 8.3.4 Experiments References 9 Underwater Descattering from Light Field 9.1 Introduction 9.2 Underwater Image Formation Model 9.2.1 Direct Radiance Term 9.2.2 Backscatter Term 9.2.3 Extension to Light Sources Outside the Medium 9.3 Single-Image Restoration 9.4 Depth Estimation 9.4.1 Defocus and Correspondence Cues 9.4.2 Transmission Depth Cue 9.4.3 Depth Fusion and Propagation 9.5 High-Quality Images from Shearing and Refocusing LF 9.6 Experiments References 10 Applications and Future Work 10.1 Application Examples of All-Weather Robot Vision 10.1.1 Unmanned Aerial Vehicle (UAV) Target Tracking Under Complex Illumination Conditions 10.1.2 Road and Navigation Line Extraction in Rural Scenes 10.1.3 Finding Passable Region for Robots from Single Image 10.1.4 Tracking in Rainy Scenes 10.1.5 Effect of Highlight Removal on Detection 10.2 Research Prospect References