Computer Vision and Recognition Systems Using Machine and Deep Learning Approaches: Fundamentals, Technologies and Applications

Title
Copyright
Contents
About the editors
Preface
1 Computer vision and recognition-based safe automated systems
	1.1 Introduction
		1.1.1 Role of computer vision in automation
		1.1.2 Organization of the chapter
	1.2 Literature survey of safe automation systems
	1.3 Application of computer vision technology in automation
		1.3.1 Using face ID in mobile devices
		1.3.2 Automated automobiles
		1.3.3 Computer vision in agriculture
		1.3.4 Computer vision in the health sector
		1.3.5 Computer vision in the e-commerce industry
		1.3.6 Generating 3D maps
		1.3.7 Classifying and detecting objects
		1.3.8 Congregation data for training algorithms
		1.3.9 Low-light mode with computer vision
	1.4 Ensuring safety during COVID-19 using computer vision
		1.4.1 AI started from bringing humans closer to forcing them in keeping apart
		1.4.2 Access control through computer vision
		1.4.3 Thermal fever detection cameras
		1.4.4 Social distancing detection
		1.4.5 Sanitization prioritization
		1.4.6 Face mask compliance
	1.5 Discussion and conclusion
	References
2 DLA: deep learning accelerator
	2.1 Introduction
	2.2 ASIC-based design accelerator
	2.3 FPGA-based design accelerator
	2.4 NoC-based design accelerator
	2.5 Flow mapping and its impact on DLAs__amp__#8217; performance
	2.6 A heuristic or dynamic algorithm__amp__#8217;s role on a DLA__amp__#8217;s efficiency
	2.7 Brief state-of-the-art survey
	References
3 Intelligent image retrieval system using deep neural networks
	3.1 Introduction
	3.2 Conventional content-based image retrieval (CBIR) system
		3.2.1 Semantic-based image retrieval (SBIR) system
	3.3 Deep learning
	3.4 Image retrieval using convolutional neural networks (CNN)
	3.5 Image retrieval using autoencoders
	3.6 Image retrieval using generative adversarial networks (GAN)
	References
4 Handwritten digits recognition using dictionary learning
	4.1 Introduction
		4.1.1 Optical character recognition
		4.1.2 Handwritten recognition
	4.2 Related works
	4.3 Dictionary learning
	4.4 DPL variants for HNR
		4.4.1 Dictionary pair learning model
		4.4.2 Incoherent dictionary pair learning (InDPL)
		4.4.3 Labeled projective dictionary pair learning
	4.5 Input data preparation
		4.5.1 Image preprocessing
		4.5.2 Histogram of oriented gradient
		4.5.3 Classification stage
	4.6 HNR datasets
	4.7 Experimental results
		4.7.1 Cross-validation
		4.7.2 Benchmarking results
	4.8 Conclusions
	References
5 Handwriting recognition using CNN and its optimization approach
	5.1 Introduction
	5.2 Related works
	5.3 Background
		5.3.1 Convolutional neural network
		5.3.2 Gated convolutional neural network
		5.3.3 Gated recurrent unit (GRU)
		5.3.4 Connectionist temporal classification (CTC)
		5.3.5 Residual operation
		5.3.6 Bi-directional gated recurrent unit (BiGRU)
		5.3.7 Squeeze and excited network (SENet)
		5.3.8 Linear bottleneck network
		5.3.9 Encoder and decoder model
	5.4 Methodology
		5.4.1 Data gathering
		5.4.2 Preprocessing
		5.4.3 Model overview
		5.4.4 Metrics
		5.4.5 Training configurations
		5.4.6 Unseen testing
		5.4.7 Inference time testing
		5.4.8 Visualize inside the model
	5.5 Experiments
		5.5.1 Experiment 1: Bluche versus Puigcerver versus Flor model
		5.5.2 Experiment 2: performance comparison of the encoder
		5.5.3 Experiment 3: performance comparison of the decoder
		5.5.4 Experiment 4: performance comparison of the skipped connection
		5.5.5 Experiment 5: performance comparison of other ResFlor model
		5.5.6 Experiment 6: ResFlor residual with SE network
		5.5.7 Experiment 7: ResFlor with residual and bottleneck network
	5.6 Summary
	5.7 Conclusion and future work
	Acknowledgments
	References
6 Real-time face mask detection on edge IoT devices
	6.1 IoT devices and object detection
		6.1.1 IoT devices and object detection
		6.1.2 Real-time object detection on edge IoT devices
		6.1.3 A generic detection algorithm
	6.2 Literature survey
	6.3 Traditional feature extraction techniques
		6.3.1 Histogram of oriented gradients (HOG)
		6.3.2 Scale invariant feature transform (SIFT)
		6.3.3 Speeded up robust features (SURF)
	6.4 Traditional detection methods
		6.4.1 Histogram of oriented gradients with support vector machines (HOG + SVM)
	6.5 Traditional face detection techniques
		6.5.1 Viola__amp__#8211;Jones Haar cascade method
	6.6 Face mask detection
	6.7 Deep learning for object detection
		6.7.1 Convolutional neural networks (CNNs)
		6.7.2 Object detection using deep learning
		6.7.3 Faster RCNN for object detection
		6.7.4 Enhancing faster RCNN with MobileNet
	6.8 Internet and deep learning
		6.8.1 Client__amp__#8211;server architecture
	6.9 Edge IoT architecture
	6.10 Implementing an edge IoT architecture
		6.10.1 Dynamic web pages
		6.10.2 Backend using Node.js
		6.10.3 MongoDB as database
	6.11 Discussion
	6.12 Conclusion
	References
7 Current challenges and applications of DeepFake systems
	7.1 Introduction to DeepFake
		7.1.1 Scenario
	7.2 Various DeepFake detection methods available and their limitations
		7.2.1 Traditional detection methods
		7.2.2 Methods based on deep learning
	7.3 Applications used to forge the multimedia
	7.4 Current challenges and future of the technology
		7.4.1 Quality of DeepFake dataset
		7.4.2 Performance evaluation
		7.4.3 Explainability of detection results
		7.4.4 Temporal aggregation
		7.4.5 Social media laundering
	7.5 Conclusion
	References
8 Vehicle control system based on eye, iris, and gesture recognition with eye tracking
	8.1 Introduction
	8.2 Eye tracking
		8.2.1 How eye tracker works
	8.3 Human gesture
		8.3.1 Head movement
	8.4 Applications of eye tracking
	8.5 Top eye tracking hardware companies
	8.6 Case studies
	8.7 Conclusion
	References
9 Sentiment analysis using deep learning
	9.1 Sentiment analysis: an interesting problem
	9.2 Sentiment and opinions
	9.3 Components of opinion
		9.3.1 Levels in sentiment analysis
		9.3.2 Classification techniques
		9.3.3 Classification types
	9.4 Deep learning
	9.5 Machine learning
	9.6 Traditional learning
	9.7 Hybrid learning approaches
	9.8 Deep neural networks
		9.8.1 Deep belief network
		9.8.2 Convolutional neural networks
		9.8.3 Stacked autoencoders
	9.9 Convolutional neural networks
		9.9.1 Word embeddings
		9.9.2 Bag of words (BOW)
		9.9.3 ConvNet structure
	9.10 Proposed model
		9.10.1 Datasets and experimental setup
		9.10.2 Results
		9.10.3 Effect of filter region size
		9.10.4 Effect of number of filters
		9.10.5 Effect of different classifiers
	9.11 Conclusions and future scope
	References
10 Classification of prefeature extracted images with deep convolutional neural network in facial emotion recognition of vehicle driver
	10.1 Introduction and related work
	10.2 Proposed models
		10.2.1 Datasets
		10.2.2 Preprocessing
		10.2.3 Prefeature extraction
		10.2.4 Convolutional neural networks
		10.2.5 Model design
		10.2.6 Metrics
		10.2.7 System configuration
	10.3 Experiments and results
	10.4 Vehicle driver emotion recognition experimental setup, results, and discussion
	10.5 Conclusion
	Acknowledgments
	References
11 MobileNet architecture and its application to computer vision
	11.1 Introduction
	11.2 Preliminaries
		11.2.1 Artificial neural network
		11.2.2 Convolution neural network
		11.2.3 Deep convolution neural network
	11.3 Benchmarked convolutional neural network
		11.3.1 VGG16
		11.3.2 Inception v3
	11.4 MobileNet architecture
		11.4.1 MobileNetv1
		11.4.2 MobileNetv2
		11.4.3 MobileNetv3
		11.4.4 NASNet mobile
	11.5 Model optimization techniques
		11.5.1 Quantization technique
	11.6 Quantized deep convolutional neural network
		11.6.1 Methodology
	11.7 Case study: healthcare domain
		11.7.1 Diabetic retinopathy
		11.7.2 Kaggle diabetic retinopathy image datasets
		11.7.3 Approach
		11.7.4 Experiment results and discussion
		11.7.5 Conclusion
	11.8 Selected MobileNet application
		11.8.1 Image classification
		11.8.2 Object detection
		11.8.3 Segmentation
	11.9 Future direction
	References
12 Study on traffic enforcement cameras monitoring to detect the wrong-way movement of vehicles using deep convolutional neural network
	12.1 Introduction
	12.2 Background
	12.3 Techniques for data collection
		12.3.1 Closed-circuit television
		12.3.2 Manual videos
	12.4 Purpose and benefit of the cameras monitoring system
	12.5 Techniques used in the monitoring of vehicles
		12.5.1 Convolution neural network
		12.5.2 R-CNN
		12.5.3 Fast region-based convolution neural network
		12.5.4 Faster R-CNN
		12.5.5 Single-shot MultiBoxDetector
		12.5.6 You Only Look Once
	12.6 Case study
		12.6.1 The detection of wrong-way drive of automobiles based on appearance using deep convolutional neural network
		12.6.2 Real-time wrong-direction detection based on deep learning
		12.6.3 A vehicle finding and counting system based on vision using deep learning
		12.6.4 A highway automobile discovery algorithm based on CNN
		12.6.5 Comparison of case studies
	12.7 Conclusion
	References
13 Glasses for smart tourism applications
	13.1 Introduction
		13.1.1 Motivation
		13.1.2 Contribution of our work
	13.2 Article structure
	13.3 Existing technologies related to smart glasses
		13.3.1 Applications of smart glasses
		13.3.2 Smart glasses technology in the market
		13.3.3 Smart glasses solutions papers
	13.4 System assumptions
	13.5 Functional architecture and technologies relevant
		13.5.1 Voice to text conversion-KALDI
		13.5.2 Name remember
		13.5.3 Facial features extraction
		13.5.4 Object (plant and animal) identification
		13.5.5 Text detection
		13.5.6 Text-translation
		13.5.7 Navigation
		13.5.8 Text to speech conversion
	13.6 Proposed style of interaction (KBSIS)
		13.6.1 KALDI
		13.6.2 Dataset and model used
		13.6.3 Inputs
		13.6.4 Inference from input and processing
		13.6.5 Outputs
	13.7 Results and discussion
		13.7.1 Navigation
		13.7.2 Music
		13.7.3 Text from image and Translate
		13.7.4 Remembering face and naming
		13.7.5 Face characteristics
		13.7.6 Weather
		13.7.7 Plant identification and search
		13.7.8 Animal identification and search
	13.8 Conclusion
	References
14 Renal calculi detection using modified grey wolf optimization
	14.1 Introduction
	14.2 Background
		14.2.1 Image segmentation
		14.2.2 Grey wolf optimization
		14.2.3 Previous work
	14.3 Proposed approach for renal calculi detection
		14.3.1 Challenges in renal calculi detection
		14.3.2 Proposed approach
	14.4 Experiment and results
		14.4.1 Dataset
		14.4.2 Performance analysis
	14.5 Conclusions and future scope
	References
15 On multi-class aerial image classification using learning machines
	15.1 Introduction
	15.2 Learning approaches
		15.2.1 Deep learning networks
		15.2.2 Feature learning
		15.2.3 Challenges for deep learning
		15.2.4 Challenges related to aerial video classification
		15.2.5 Applications
	15.3 Learning architecture and classification
		15.3.1 Supervised learning architectures
		15.3.2 Unsupervised learning
		15.3.3 Deep learning for planning and situational awareness
		15.3.4 Deep learning for motion control
		15.3.5 Object detection
		15.3.6 Classification
	15.4 Training
		15.4.1 Weight initialization
		15.4.2 Convolutional methods
		15.4.3 Activation functions
		15.4.4 Subsampling or pooling layer
		15.4.5 Optimization techniques
		15.4.6 Benchmark datasets
	15.5 Energy efficiency in learning approaches
	15.6 Performance metrics
	15.7 Development kits and frameworks
	15.8 Discussions and future directions
	References
16 Machine learning methodology toward identification of mature citrus fruits
	16.1 Introduction
		16.1.1 Harvesting
		16.1.2 Farm automation
		16.1.3 Fruit detection
		16.1.4 Proposed method
	16.2 Literature survey
	16.3 Implementation
		16.3.1 Image acquisition and data collection
		16.3.2 Pre-processing
		16.3.3 Feature extraction
		16.3.4 Machine learning model and database formation
		16.3.5 Data retrieval
		16.3.6 Match with dataset or testing the model
		16.3.7 Display result
		16.3.8 Application design
	16.4 Experiments and result
		16.4.1 Qualification measures
		16.4.2 Training result
		16.4.3 Testing result
	16.5 Conclusion
		16.5.1 Future scope
	References
17 Automated detection of defects and grading of cashew kernels using machine learning
	17.1 Introduction
		17.1.1 Related work
		17.1.2 Proposed methodology
	17.2 Defects and grades of cashew kernels
		17.2.1 Cashew kernel manufacturing process
		17.2.2 Defects of cashew kernel
		17.2.3 Grades of cashew kernel
	17.3 Implementation of the methodology
		17.3.1 Image preprocessing and segmentation
		17.3.2 Feature extraction
		17.3.3 Classification
	17.4 Results and discussions
	17.5 Conclusion
	References
Index