Laser Scanning Systems in Highway and Safety Assessment: Analysis of Highway Geometry and Safety Using LiDAR (Advances in Science, Technology & Innovation)

"This book aims to promote the core understanding of a proper modelling of road traffic accidents by deep learning methods using traffic information and road geometry delineated from laser scanning data. The first two chapters of the book introduce the reader to laser scanning technology with creative explanation and graphical illustrations, review and recent methods of extracting geometric road parameters. The next three chapters present different machine learning and statistical techniques applied to extract road geometry information from laser scanning data. Chapters 6 and 7 present methods for modelling roadside features and automatic road geometry identification in vector data. After that, this book goes on reviewing methods used for road traffic accident modelling including accident frequency and injury severity of the traffic accident (Chapter 8). Then, the next chapter explores the details of neural networks and their performance in predicting the traffic accidents along with a comparison with common data mining models. Chapter 10 presents a novel hybrid model combining extreme gradient boosting and deep neural networks for predicting injury severity of road traffic accidents. This chapter is followed by deep learning applications in modelling accident data using feed-forward, convolutional, recurrent neural network models (Chapter 11). The final chapter (Chapter 12) presents a procedure for modelling traffic accident with little data based on the concept of transfer learning. This book aims to help graduate students, professionals, decision makers, and road planners in developing better traffic accident prediction models using advanced neural networks."-- Provided by publisher Preface 6 Contents 12 Road Geometry Modelling 17 1 Laser Scanning Technologies in Road Geometry Modeling 18 1.1 Background 18 1.2 Laser Scanning Systems 18 1.2.1 Airborne Laser Scanning 18 1.2.1.1 Height Features of ALS Data 19 1.2.1.2 Laser Intensity 19 1.2.1.3 Aerial Orthophotos 20 1.2.1.4 Specifications of ALS Sensors in Road Geometric Modeling 20 1.2.2 Mobile Laser Scanning 21 1.2.3 Terrestrial Laser Scanning 22 1.3 Description of Scanner Field of View 23 1.4 Comparison of ALS, MLS, and TLS for Road Extraction and Modeling 23 1.5 Conclusion 26 References 26 2 Road Geometric Modeling Using Laser Scanning Data: A Critical Review 29 2.1 Introduction 29 2.2 Road Geometric Model 29 2.3 Delineation of Road Geometric Information: A Novel Classification for LiDAR Data 31 2.4 Road Extraction Based on Types of Sensors 31 2.4.1 Satellite Images 31 2.4.2 Aerial Photographs and Unmanned Aerial Vehicles (UAV) 31 2.4.3 LiDAR 32 2.5 Classification of Road Extraction According to the Preset Objective 32 2.5.1 Road Centerline Extraction 32 2.5.2 Road Extraction in Two Dimensions 33 2.5.3 Road Reconstruction and Geometric Modeling 34 2.6 Classification of Road Extraction According to the Technique Applied 34 2.6.1 Unsupervised/Supervised Classification 34 2.6.2 Machine Learning 35 2.6.3 Object-Based Image Analysis (OBIA) 36 2.6.4 Morphology and Filtrate 37 2.6.5 Active Contour Models 37 2.6.6 Information Fusion 38 2.6.7 Hierarchal and Hybrid Methods 38 2.7 Discussion and Current Challenges 39 2.8 Conclusion 42 References 42 3 Road Geometric Modeling Using a Novel Hierarchical Approach 46 3.1 Introduction 46 3.2 Previous Works 46 3.3 Methodology 47 3.3.1 MS Segmentation 48 3.3.2 Optimization of Segmentation Parameters 49 3.3.3 Classification of Segments Using SVM 50 3.3.4 Extraction of Geometric Road Parameters Using PCA Transformation 50 3.4 Experimental Results 50 3.5 Accuracy Assessment 55 3.6 Evaluation of Model Transferability 55 3.7 Discussion 57 3.8 Conclusion 57 References 58 4 Optimizing Support Vector Machine and Ensemble Trees Using Taguchi Method for Road Extraction from LiDAR Data 60 4.1 Introduction 60 4.1.1 SVM 61 4.1.2 Ensemble Trees 61 4.1.3 Taguchi-Based Optimization of User-Defined Parameters 62 4.2 Study Area 62 4.3 Methods 63 4.3.1 Overall Workflow of the Study 63 4.3.2 Data Preprocessing and Preparation 63 4.3.3 Optimization of User-Defined Parameters (Taguchi Method) 64 4.3.4 Training the SVM and Ensemble Tree Algorithms with Default and Optimized Parameters 64 4.4 Results and Discussion 65 4.4.1 Results of the Classification Using the Default User-Defined Parameters 65 4.4.2 Results of the Taguchi-Based Optimization 66 4.4.2.1 Optimization of the SVM Parameters 66 4.4.2.2 Optimization of the Ensemble Tree Parameters 67 4.4.2.3 Analysis of Variance (ANOVA) 67 4.4.3 Results of the Classification Using the Optimized User-Defined Parameters 69 4.4.4 Accuracy and Performance Assessments 69 4.5 Conclusion 71 References 71 5 An Integrated Machine Learning Approach for Automatic Highway Extraction from Airborne LiDAR Data and Orthophotos 74 5.1 Introduction 74 5.2 Previous Related Works 75 5.3 Machine Learning Models 76 5.3.1 Multilayer Perceptron Neural Networks (MLP) 76 5.3.2 Support Vector Machine (SVM) 76 5.3.3 Logistic Regression (LR) 77 5.3.4 Decision Tree (DT) 77 5.4 Study Area 78 5.5 Data and Methodology 78 5.5.1 Data Preprocessing 78 5.5.1.1 Generation of Digital Elevation Model (DEM) 78 5.5.1.2 Color Space Transformation 79 5.5.2 Preparation of Input Attributes and Training/Testing Samples 79 5.5.3 Proposed GIS Workflow for Automatic Highway Extraction 79 5.6 Results and Discussion 80 5.6.1 Proposed Models for Highway Extraction 81 5.6.2 Accuracy Assessment 81 5.6.3 Multilayer Perceptron 81 5.6.4 Support Vector Machine 83 5.6.5 Applications on Raster Data and Models Transferability Issues 83 5.6.6 Quantitative Evaluation of Road Extraction 84 5.7 Conclusion 86 References 87 6 Effect of Roadside Features on Injury Severity of Traffic Accidents 90 6.1 Introduction 90 6.2 Methods 91 6.2.1 Overall Flowchart 91 6.2.2 Data Preparation and Preprocessing 91 6.2.3 Roadside Feature Extraction 91 6.2.3.1 Segmentation 91 6.2.3.2 Feature Extraction 91 6.2.3.3 Classification 93 6.2.4 Feature Ranking 94 6.2.5 Impact Analysis 94 6.3 Experimental Results 94 6.3.1 Dataset 94 6.3.2 Roadside Features 94 6.3.3 Results of LR Modeling 96 6.3.4 Results of Impact Assessment 96 6.3.4.1 Height Features of ALS Data Effects of Road Median, Width, and Shoulder Width Features 96 6.3.4.2 Effects of Lighting Conditions and Tree Density 96 6.3.4.3 Effects of Vehicle Types Involved in Accidents 96 6.3.5 Results of Feature Contributions 97 6.4 Conclusion 97 References 98 7 Novel GIS-Based Model for Automatic Identification of Road Geometry in Vector Data 100 7.1 Introduction 100 7.2 Methods 100 7.2.1 Airborne Laser Scanning 100 7.2.2 Proposed Model 101 7.2.3 Extraction of Road Geometric Design Parameters 101 7.3 Results and Discussions 102 7.3.1 Application of Bezier Interpolation on Road Polylines 102 7.3.2 Results of Road Geometry Identified in Vector Data 103 7.3.3 Characteristics of Identified Curves 104 7.3.4 Validation of Different Classification Methods 104 7.3.5 Results of Calculating NSE Design Parameters 104 7.4 Conclusion 107 References 107 Modeling Road Traffic Accidents 108 8 Review of Traffic Accident Predictions with Neural Networks 109 8.1 Introduction 109 8.2 Background of Neural Networks 109 8.3 The Common Types of Neural Networks 110 8.3.1 Feedforward Neural Networks (FNNs) 110 8.3.2 Convolutional Neural Network (CNN) 111 8.3.3 Recurrent Neural Network (RNN) 112 8.4 Road Accident Setting 113 8.5 Modeling Road Traffic Accidents 113 8.5.1 Model Factors for Predicting Road Accident Frequency 113 8.5.2 Model Factors for Predicting Road Accident Severity 114 8.5.3 Frequency Modeling 114 8.6 Injury Severity Modeling 115 8.7 The Use of Neural Networks for Road Traffic Accident Modeling 116 8.8 The Rationale of Using NN 116 8.9 Comparison of NN and Statistical Models for Traffic Accident Modeling 116 8.10 Predictive Performance of Different Types of NN 119 8.11 Conclusion 120 References 120 9 Modeling Traffic Accident Severity Using Neural Networks and Support Vector Machines 122 9.1 Introduction 122 9.2 Materials and Methods 122 9.3 Study Area and Data 123 9.3.1 DNNs 123 9.3.2 SVM 123 9.4 Results and Discussion 124 9.4.1 Results of DNN Model 124 9.4.2 Results of SVM Model 125 9.4.3 Results of RF Factor Importance 125 9.5 Conclusion 127 References 127 10 Predicting Injury Severity of Road Traffic Accidents Using a Hybrid Extreme Gradient Boosting and Deep Neural Network Approach 129 10.1 Introduction 129 10.2 Previous Works 129 10.3 Methodology 130 10.3.1 Data and Application Site 130 10.3.2 Model Development 130 10.3.2.1 Overview 130 10.3.2.2 Handling Imbalanced Data 131 10.3.2.3 Feature Engineering and Selection 132 10.3.2.4 Hybrid Model 132 10.3.3 Evaluation Method 132 10.4 Results and Discussion 133 10.4.1 Results of Handling Imbalanced Data 133 10.4.2 Results of Feature Importance 134 10.4.3 Performance of Hybrid Model 134 10.4.4 Comparison with Other Methods 134 10.5 Conclusion 136 References 137 11 Applications of Deep Learning in Severity Prediction of Traffic Accidents 138 11.1 Introduction 138 11.2 Deep Learning Models 138 11.2.1 Feedforward NNs 138 11.2.2 CNNs 139 11.2.3 RNNs 140 11.3 Proposed Models 140 11.4 Experimental Results 141 11.4.1 Dataset 142 11.4.2 Model Performance 142 11.4.3 Optimization and Sensitivity Analysis 142 11.5 Discussion 146 11.6 Conclusion 147 References 147 12 Forecasting Severity of Motorcycle Crashes Using Transfer Learning 149 12.1 Introduction 149 12.2 RNN 150 12.3 TL 150 12.4 Proposed Method 151 12.4.1 Network Architecture 152 12.4.2 Network Parameters 152 12.4.3 Data Transformation 153 12.4.4 Hyperparameter Optimization 153 12.4.5 Mitigating Overfitting Problem 154 12.5 Experimental Results 154 12.5.1 Dataset 154 12.5.2 Training from Scratch Versus TL 157 12.5.3 Effects of Batch Size 159 12.5.4 Effects of Data Transformation 160 12.5.5 Effects of Network Hyperparameters 160 12.6 Discussion 161 12.6.1 Time Complexity of the Model 161 12.6.2 Model Comparisons 162 12.6.3 Importance of Accident-Related Factors 162 12.7 Conclusion 163 References 164 Front Matter ....Pages i-xv Front Matter ....Pages 1-1 Laser Scanning Technologies in Road Geometry Modeling (Biswajeet Pradhan, Maher Ibrahim Sameen)....Pages 3-13 Road Geometric Modeling Using Laser Scanning Data: A Critical Review (Biswajeet Pradhan, Maher Ibrahim Sameen)....Pages 15-31 Road Geometric Modeling Using a Novel Hierarchical Approach (Biswajeet Pradhan, Maher Ibrahim Sameen)....Pages 33-46 Optimizing Support Vector Machine and Ensemble Trees Using Taguchi Method for Road Extraction from LiDAR Data (Biswajeet Pradhan, Maher Ibrahim Sameen)....Pages 47-60 An Integrated Machine Learning Approach for Automatic Highway Extraction from Airborne LiDAR Data and Orthophotos (Biswajeet Pradhan, Maher Ibrahim Sameen)....Pages 61-76 Effect of Roadside Features on Injury Severity of Traffic Accidents (Biswajeet Pradhan, Maher Ibrahim Sameen)....Pages 77-86 Novel GIS-Based Model for Automatic Identification of Road Geometry in Vector Data (Biswajeet Pradhan, Maher Ibrahim Sameen)....Pages 87-94 Front Matter ....Pages 95-95 Review of Traffic Accident Predictions with Neural Networks (Biswajeet Pradhan, Maher Ibrahim Sameen)....Pages 97-109 Modeling Traffic Accident Severity Using Neural Networks and Support Vector Machines (Biswajeet Pradhan, Maher Ibrahim Sameen)....Pages 111-117 Predicting Injury Severity of Road Traffic Accidents Using a Hybrid Extreme Gradient Boosting and Deep Neural Network Approach (Biswajeet Pradhan, Maher Ibrahim Sameen)....Pages 119-127 Applications of Deep Learning in Severity Prediction of Traffic Accidents (Biswajeet Pradhan, Maher Ibrahim Sameen)....Pages 129-139 Forecasting Severity of Motorcycle Crashes Using Transfer Learning (Biswajeet Pradhan, Maher Ibrahim Sameen)....Pages 141-157