Computational Intelligence Based Solutions for Vision Systems

Computer vision and image processing-based systems and their applications are already an integral part of modern living and are expected to increase in prevalence and complexity. Vision system provides the ability to handle and examine the large data generated by cameras and make a decision based on the situational requirement. As computational intelligent methods are especially adept at rapidly resolving inexact situations or where there is incomplete knowledge, they are being heavily researched and employed in this space. This merger creates intelligent vision systems, which can be extremely versatile, and this book focusses on the latest developments and current key research areas in the field. Key Features: Interdisciplinary approach to intelligent computing applications for machine vision Encompasses high performance computing for vision systems and control Includes present applications and challenges for future development Reviews range of CI and ML methodologies International author pool PRELIMS.pdf Preface Acknowledgements Editors biographies Varun Bajaj Irshad Ahmad Ansari List of contributors CH001.pdf Chapter 1 Drone-based vision system: surveillance during calamities 1.1 Introduction 1.2 Surveillance system 1.2.1 The importance of surveillance systems 1.2.2 The use of drones in surveillance system 1.3 Proposed method 1.3.1 Detecting human faces 1.3.2 Tracking human faces 1.3.3 Locating and capturing human faces 1.3.4 Counting the number of people 1.3.5 Drone deployment and testing 1.4 Conclusion Acknowledgements References CH002.pdf Chapter 2 Use of computer vision to inspect automatically machined workpieces 2.1 Introduction 2.2 Related works 2.3 Methods 2.3.1 Image acquisition 2.3.2 Surface analysis to determine workpiece quality 2.3.3 Burr detection 2.3.4 Classification 2.4 Experimental set-up 2.5 Experimental results 2.5.1 Workpiece quality 2.5.2 Burrs 2.6 Conclusions and future work Acknowledgements References CH003.pdf Chapter 3 Machine learning for vision based crowd management 3.1 Introduction 3.2 Related work 3.2.1 A review of people count detection techniques 3.3 Proposed methodology 3.3.1 The architecture of the proposed system 3.3.2 An objective technique for counting people 3.3.3 The architecture of YOLOV3 3.4 Experimental results 3.4.1 Dataset 3.4.2 Performance analysis 3.5 Conclusion References CH004.pdf Chapter 4 Skin cancer classification model based on hybrid deep feature generation and iterative mRMR 4.1 Introduction 4.1.1 Background 4.1.2 Motivation 4.1.3 Literature review 4.1.4 Our model 4.1.5 Contributions 4.1.6 Study outline 4.2 Material 4.3 Preliminary 4.3.1 Residual networks 4.3.2 DenseNet201 model 4.3.3 MobileNetV2 model 4.3.4 ShuffleNet model 4.4 The proposed framework 4.4.1 Feature generation 4.4.2 Iterative mRMR feature selector 4.4.3 Classification 4.5 Results and discussion 4.5.1 Experimental set-up 4.5.2 Results 4.5.3 Discussion 4.6 Conclusions and future works References CH005.pdf Chapter 5 An analysis of human activity recognition systems and their importance in the current era 5.1 Introduction 5.2 Stages in human activity recognition 5.3 Applications of human activity recognition 5.3.1 Security video surveillance and home monitoring 5.3.2 Retail 5.3.3 Healthcare 5.3.4 Smart homes 5.3.5 Workplace monitoring 5.3.6 Entertainment 5.4 Approaches for human activity recognition 5.4.1 The HAR process using 3D posture data 5.4.2 Human action recognition using DFT 5.4.3 The local SVM approach 5.4.4 A robust approach for action recognition based on spatio-temporal features in RGB-D sequences 5.4.5 SlowFast networks for video recognition 5.4.6 Long-term recurrent convolutional networks for visual recognition and description 5.4.7 3D convolutional neural networks for human action recognition 5.4.8 Human activity recognition using an optical flow based feature set 5.4.9 Learning a hierarchical spatio-temporal model 5.4.10 Human action recognition using trajectory-based representation 5.4.11 Human activity recognition using a deep neural network with contextual information 5.5 Challenges in human activity recognition 5.5.1 Dataset 5.5.2 Sensors 5.5.3 Experimentation environment 5.5.4 Intraclass variation and interclass similarity 5.5.5 Multi-subject interactions and group activities 5.5.6 Training 5.5.7 Challenges in HAR applications 5.6 Datasets available for activity detection research 5.6.1 Action-level dataset 5.6.2 Interaction-level dataset 5.6.3 Group activities level dataset 5.6.4 Behavior-level dataset 5.7 Scope for further research in this domain 5.8 Conclusion References CH006.pdf Chapter 6 A deep learning-based food detection and classification system 6.1 Introduction 6.2 Literature review 6.3 Theory 6.3.1 YOLOv3 6.3.2 YOLOv4 6.3.3 SSD 6.4 Methodology/experiments 6.4.1 Dataset 6.4.2 Data augmentation 6.4.3 Implementation 6.4.4 Software and hardware 6.4.5 Performance parameters 6.5 Results 6.6 Conclusion and future scope References CH007.pdf Chapter 7 The detection of images recaptured through screenshots based on spatial rich model analysis 7.1 Introduction 7.2 Literature review 7.3 Spatial rich model 7.3.1 Computing noise residuals 7.3.2 Residual truncation and quantization 7.3.3 Formation of a sub-model with co-occurrence matrices 7.4 Proposed work 7.4.1 Selection of the neighborhood descriptor 7.5 Experimental results 7.5.1 Screenshot dataset 7.5.2 Detection performance of the neighborhood descriptors 7.5.3 The detection performance of neighborhood descriptors with an ensemble classifier 7.5.4 Detection performance of neighborhood descriptors with an SVM 7.5.5 Performance comparison of the neighborhood descriptors 7.6 Conclusion 7.7 Future work Acknowledgements References CH008.pdf Chapter 8 Data augmentation for deep ensembles in polyp segmentation 8.1 Introduction 8.2 Deep learning for semantic image segmentation 8.3 Stochastic activation selection 8.4 Data augmentation 8.4.1 Spatial stretch 8.4.2 Shadows 8.4.3 Contrast and motion blur 8.4.4 Color change and rotation 8.4.5 Segmentation 8.4.6 Rand augment 8.4.7 RICAP 8.4.8 Color and shape change 8.4.9 Occlusion 1 8.4.10 Occlusion 2 8.4.11 GridMask 8.4.12 AttentiveCutMix 8.4.13 Modified ResizeMix 8.4.14 Color mapping 8.5 Results on colorectal cancer segmentation 8.5.1 Datasets, testing protocol and metrics 8.5.2 Experiments 8.6 Conclusion Acknowledgments References CH009.pdf Chapter 9 Identification of the onset of Parkinson’s disease through a multiscale classification deep learning model utilizing a fusion of multiple conventional features with an nDS spatially exploited symmetrical convolutional pattern 9.1 Introduction 9.1.1. A comprehensive literature review 9.1.2 Contributions 9.2 Proposed methodology 9.2.1 Retrieval of voice samples 9.2.2 Pre-processing 9.2.3 Proposed multiscale multiple feature convolution with hybrid n-dilations (MMFCHnD) architecture 9.3 Experimental results and discussion 9.3.1 Evaluation metrics 9.3.2 Development of the training and testing images 9.3.3 Deep learning training details 9.3.4 Implementation results 9.4 Conclusion References CH010.pdf Chapter 10 Computer vision approach with deep learning for a medical intelligence system 10.1 Introduction 10.2 Defining computer vision 10.3 Computer vision in practice 10.3.1 Medical imaging 10.3.2 Cardiology 10.3.3 Pathology 10.3.4 Dermatology 10.3.5 Ophthalmology 10.3.6 Video for medical purposes 10.3.7 The presence of humans 10.3.8 Implementation in the clinic 10.4 A case study of vision based machine learning 10.4.1 Networks of neurons 10.5 Data preparation overview 10.5.1 Data access and querying 10.5.2 De-identification 10.5.3 Data retention 10.5.4 Medical image resembling 10.5.5 Choosing an appropriate label and a definition of ground truth 10.5.6 The truth or the label’s quality 10.6 The future of computer vision and natural language processing in healthcare 10.7 Research related problems in computer vision 10.7.1 View of CNN through computer vision 10.7.2 Visualizations based on gradients References CH011.pdf Chapter 11 Machine learning in medicine: diagnosis of skin cancer using a support vector machine (SVM) classifier 11.1 Introduction 11.2 Technologies used in skin cancer detection 11.3 Support vector machines (SVMs) 11.4 The SVM in skin cancer detection 11.4.1 Image acquisition 11.4.2 Feature extraction 11.4.3 SVM classification 11.5 Brief description of skin cancer detection 11.6 Challenges faced by SVMs 11.7 Future aspects in skin cancer detection 11.8 Conclusion References