Enhancing licence plate recognition for a robust vehicle re-identification system
- Authors: Boby, Alden Zachary
- Date: 2024-10-11
- Subjects: Automobile theft South Africa , Deep learning (Machine learning) , Object detection , YOLOv7 , YOLO , Pattern recognition systems , Image processing Digital techniques , Automobile license plates
- Language: English
- Type: Academic theses , Master's theses , text
- Identifier: http://hdl.handle.net/10962/464322 , vital:76501
- Description: Vehicle security is a growing concern for citizens of South Africa. Law enforcement relies on reports and security camera footage for vehicle identification but struggles to match the increasing number of carjacking incidents and low vehicle recovery rates. Security camera footage offers an accessible means to identify stolen vehicles, yet it often poses hurdles like anamorphic plates and low resolution. Furthermore, depending on human operators proves inefficient, requiring faster processes to improve vehicle recovery rates and trust in law enforcement. The integration of deep learning has revolutionised object detection algorithms, increasing the popularity of vehicle tracking for security purposes. This thesis investigates advanced deep-learning methods for a comprehensive vehicle search and re-identification system. It enhances YOLOv7’s algorithmic capabilities and employs preprocessing techniques like super-resolution and perspective correction via the Improved Warped Planar Object Detection network for more effective licence plate optical character recognition. Key contributions include a specifically annotated dataset for training object detection models, an optical character recognition model based on YOLOv7, and a method for identifying vehicles in unrestricted data. The system detected rectangular and square licence plates without prior shape knowledge, achieving a 98.7% character recognition rate compared to 95.31% in related work. Moreover, it outperformed traditional optical character recognition by 28.25% and deep-learning EasyOCR by 14.18%. Its potential applications in law enforcement, traffic management, and parking systems can improve surveillance and security through automation. , Thesis (MSc) -- Faculty of Science, Computer Science, 2024
- Full Text:
- Date Issued: 2024-10-11
- Authors: Boby, Alden Zachary
- Date: 2024-10-11
- Subjects: Automobile theft South Africa , Deep learning (Machine learning) , Object detection , YOLOv7 , YOLO , Pattern recognition systems , Image processing Digital techniques , Automobile license plates
- Language: English
- Type: Academic theses , Master's theses , text
- Identifier: http://hdl.handle.net/10962/464322 , vital:76501
- Description: Vehicle security is a growing concern for citizens of South Africa. Law enforcement relies on reports and security camera footage for vehicle identification but struggles to match the increasing number of carjacking incidents and low vehicle recovery rates. Security camera footage offers an accessible means to identify stolen vehicles, yet it often poses hurdles like anamorphic plates and low resolution. Furthermore, depending on human operators proves inefficient, requiring faster processes to improve vehicle recovery rates and trust in law enforcement. The integration of deep learning has revolutionised object detection algorithms, increasing the popularity of vehicle tracking for security purposes. This thesis investigates advanced deep-learning methods for a comprehensive vehicle search and re-identification system. It enhances YOLOv7’s algorithmic capabilities and employs preprocessing techniques like super-resolution and perspective correction via the Improved Warped Planar Object Detection network for more effective licence plate optical character recognition. Key contributions include a specifically annotated dataset for training object detection models, an optical character recognition model based on YOLOv7, and a method for identifying vehicles in unrestricted data. The system detected rectangular and square licence plates without prior shape knowledge, achieving a 98.7% character recognition rate compared to 95.31% in related work. Moreover, it outperformed traditional optical character recognition by 28.25% and deep-learning EasyOCR by 14.18%. Its potential applications in law enforcement, traffic management, and parking systems can improve surveillance and security through automation. , Thesis (MSc) -- Faculty of Science, Computer Science, 2024
- Full Text:
- Date Issued: 2024-10-11
Investigating unimodal isolated signer-independent sign language recognition
- Authors: Marais, Marc Jason
- Date: 2024-04-04
- Subjects: Convolutional neural network , Sign language recognition , Human activity recognition , Pattern recognition systems , Neural networks (Computer science)
- Language: English
- Type: Academic theses , Master's theses , text
- Identifier: http://hdl.handle.net/10962/435343 , vital:73149
- Description: Sign language serves as the mode of communication for the Deaf and Hard of Hearing community, embodying a rich linguistic and cultural heritage. Recent Sign Language Recognition (SLR) system developments aim to facilitate seamless communication between the Deaf community and the broader society. However, most existing systems are limited by signer-dependent models, hindering their adaptability to diverse signing styles and signers, thus impeding their practical implementation in real-world scenarios. This research explores various unimodal approaches, both pose-based and vision-based, for isolated signer-independent SLR using RGB video input on the LSA64 and AUTSL datasets. The unimodal RGB-only input strategy provides a realistic SLR setting where alternative data sources are either unavailable or necessitate specialised equipment. Through systematic testing scenarios, isolated signer-independent SLR experiments are conducted on both datasets, primarily focusing on AUTSL – a signer-independent dataset. The vision-based R(2+1)D-18 model emerged as the top performer, achieving 90.64% accuracy on the unseen AUTSL dataset test split, closely followed by the pose-based Spatio- Temporal Graph Convolutional Network (ST-GCN) model with an accuracy of 89.95%. Furthermore, these models achieved comparable accuracies at a significantly lower computational demand. Notably, the pose-based approach demonstrates robust generalisation to substantial background and signer variation. Moreover, the pose-based approach demands significantly less computational power and training time than vision-based approaches. The proposed unimodal pose-based and vision-based systems were concluded to both be effective at classifying sign classes in the LSA64 and AUTSL datasets. , Thesis (MSc) -- Faculty of Science, Ichthyology and Fisheries Science, 2024
- Full Text:
- Date Issued: 2024-04-04
- Authors: Marais, Marc Jason
- Date: 2024-04-04
- Subjects: Convolutional neural network , Sign language recognition , Human activity recognition , Pattern recognition systems , Neural networks (Computer science)
- Language: English
- Type: Academic theses , Master's theses , text
- Identifier: http://hdl.handle.net/10962/435343 , vital:73149
- Description: Sign language serves as the mode of communication for the Deaf and Hard of Hearing community, embodying a rich linguistic and cultural heritage. Recent Sign Language Recognition (SLR) system developments aim to facilitate seamless communication between the Deaf community and the broader society. However, most existing systems are limited by signer-dependent models, hindering their adaptability to diverse signing styles and signers, thus impeding their practical implementation in real-world scenarios. This research explores various unimodal approaches, both pose-based and vision-based, for isolated signer-independent SLR using RGB video input on the LSA64 and AUTSL datasets. The unimodal RGB-only input strategy provides a realistic SLR setting where alternative data sources are either unavailable or necessitate specialised equipment. Through systematic testing scenarios, isolated signer-independent SLR experiments are conducted on both datasets, primarily focusing on AUTSL – a signer-independent dataset. The vision-based R(2+1)D-18 model emerged as the top performer, achieving 90.64% accuracy on the unseen AUTSL dataset test split, closely followed by the pose-based Spatio- Temporal Graph Convolutional Network (ST-GCN) model with an accuracy of 89.95%. Furthermore, these models achieved comparable accuracies at a significantly lower computational demand. Notably, the pose-based approach demonstrates robust generalisation to substantial background and signer variation. Moreover, the pose-based approach demands significantly less computational power and training time than vision-based approaches. The proposed unimodal pose-based and vision-based systems were concluded to both be effective at classifying sign classes in the LSA64 and AUTSL datasets. , Thesis (MSc) -- Faculty of Science, Ichthyology and Fisheries Science, 2024
- Full Text:
- Date Issued: 2024-04-04
A multispectral and machine learning approach to early stress classification in plants
- Authors: Poole, Louise Carmen
- Date: 2022-04-06
- Subjects: Machine learning , Neural networks (Computer science) , Multispectral imaging , Image processing , Plant stress detection
- Language: English
- Type: Master's thesis , text
- Identifier: http://hdl.handle.net/10962/232410 , vital:49989
- Description: Crop loss and failure can impact both a country’s economy and food security, often to devastating effects. As such, the importance of successfully detecting plant stresses early in their development is essential to minimize spread and damage to crop production. Identification of the stress and the stress-causing agent is the most critical and challenging step in plant and crop protection. With the development of and increase in ease of access to new equipment and technology in recent years, the use of spectroscopy in the early detection of plant diseases has become notably popular. This thesis narrows down the most suitable multispectral imaging techniques and machine learning algorithms for early stress detection. Datasets were collected of visible images and multispectral images. Dehydration was selected as the plant stress type for the main experiments, and data was collected from six plant species typically used in agriculture. Key contributions of this thesis include multispectral and visible datasets showing plant dehydration as well as a separate preliminary dataset on plant disease. Promising results on dehydration showed statistically significant accuracy improvements in the multispectral imaging compared to visible imaging for early stress detection, with multispectral input obtaining a 92.50% accuracy over visible input’s 77.50% on general plant species. The system was effective at stress detection on known plant species, with multispectral imaging introducing greater improvement to early stress detection than advanced stress detection. Furthermore, strong species discrimination was achieved when exclusively testing either early or advanced dehydration against healthy species. , Thesis (MSc) -- Faculty of Science, Ichthyology & Fisheries Sciences, 2022
- Full Text:
- Date Issued: 2022-04-06
- Authors: Poole, Louise Carmen
- Date: 2022-04-06
- Subjects: Machine learning , Neural networks (Computer science) , Multispectral imaging , Image processing , Plant stress detection
- Language: English
- Type: Master's thesis , text
- Identifier: http://hdl.handle.net/10962/232410 , vital:49989
- Description: Crop loss and failure can impact both a country’s economy and food security, often to devastating effects. As such, the importance of successfully detecting plant stresses early in their development is essential to minimize spread and damage to crop production. Identification of the stress and the stress-causing agent is the most critical and challenging step in plant and crop protection. With the development of and increase in ease of access to new equipment and technology in recent years, the use of spectroscopy in the early detection of plant diseases has become notably popular. This thesis narrows down the most suitable multispectral imaging techniques and machine learning algorithms for early stress detection. Datasets were collected of visible images and multispectral images. Dehydration was selected as the plant stress type for the main experiments, and data was collected from six plant species typically used in agriculture. Key contributions of this thesis include multispectral and visible datasets showing plant dehydration as well as a separate preliminary dataset on plant disease. Promising results on dehydration showed statistically significant accuracy improvements in the multispectral imaging compared to visible imaging for early stress detection, with multispectral input obtaining a 92.50% accuracy over visible input’s 77.50% on general plant species. The system was effective at stress detection on known plant species, with multispectral imaging introducing greater improvement to early stress detection than advanced stress detection. Furthermore, strong species discrimination was achieved when exclusively testing either early or advanced dehydration against healthy species. , Thesis (MSc) -- Faculty of Science, Ichthyology & Fisheries Sciences, 2022
- Full Text:
- Date Issued: 2022-04-06
Investigating combinations of feature extraction and classification for improved image-based multimodal biometric systems at the feature level
- Authors: Brown, Dane L
- Date: 2018
- Language: English
- Type: text , Thesis , Doctoral , PhD
- Identifier: http://hdl.handle.net/10962/63470 , vital:28414
- Description: Multimodal biometrics has become a popular means of overcoming the limitations of unimodal biometric systems. However, the rich information particular to the feature level is of a complex nature and leveraging its potential without overfitting a classifier is not well studied. This research investigates feature-classifier combinations on the fingerprint, face, palmprint, and iris modalities to effectively fuse their feature vectors for a complementary result. The effects of different feature-classifier combinations are thus isolated to identify novel or improved algorithms. A new face segmentation algorithm is shown to increase consistency in nominal and extreme scenarios. Moreover, two novel feature extraction techniques demonstrate better adaptation to dynamic lighting conditions, while reducing feature dimensionality to the benefit of classifiers. A comprehensive set of unimodal experiments are carried out to evaluate both verification and identification performance on a variety of datasets using four classifiers, namely Eigen, Fisher, Local Binary Pattern Histogram and linear Support Vector Machine on various feature extraction methods. The recognition performance of the proposed algorithms are shown to outperform the vast majority of related studies, when using the same dataset under the same test conditions. In the unimodal comparisons presented, the proposed approaches outperform existing systems even when given a handicap such as fewer training samples or data with a greater number of classes. A separate comprehensive set of experiments on feature fusion show that combining modality data provides a substantial increase in accuracy, with only a few exceptions that occur when differences in the image data quality of two modalities are substantial. However, when two poor quality datasets are fused, noticeable gains in recognition performance are realized when using the novel feature extraction approach. Finally, feature-fusion guidelines are proposed to provide the necessary insight to leverage the rich information effectively when fusing multiple biometric modalities at the feature level. These guidelines serve as the foundation to better understand and construct biometric systems that are effective in a variety of applications.
- Full Text:
- Date Issued: 2018
- Authors: Brown, Dane L
- Date: 2018
- Language: English
- Type: text , Thesis , Doctoral , PhD
- Identifier: http://hdl.handle.net/10962/63470 , vital:28414
- Description: Multimodal biometrics has become a popular means of overcoming the limitations of unimodal biometric systems. However, the rich information particular to the feature level is of a complex nature and leveraging its potential without overfitting a classifier is not well studied. This research investigates feature-classifier combinations on the fingerprint, face, palmprint, and iris modalities to effectively fuse their feature vectors for a complementary result. The effects of different feature-classifier combinations are thus isolated to identify novel or improved algorithms. A new face segmentation algorithm is shown to increase consistency in nominal and extreme scenarios. Moreover, two novel feature extraction techniques demonstrate better adaptation to dynamic lighting conditions, while reducing feature dimensionality to the benefit of classifiers. A comprehensive set of unimodal experiments are carried out to evaluate both verification and identification performance on a variety of datasets using four classifiers, namely Eigen, Fisher, Local Binary Pattern Histogram and linear Support Vector Machine on various feature extraction methods. The recognition performance of the proposed algorithms are shown to outperform the vast majority of related studies, when using the same dataset under the same test conditions. In the unimodal comparisons presented, the proposed approaches outperform existing systems even when given a handicap such as fewer training samples or data with a greater number of classes. A separate comprehensive set of experiments on feature fusion show that combining modality data provides a substantial increase in accuracy, with only a few exceptions that occur when differences in the image data quality of two modalities are substantial. However, when two poor quality datasets are fused, noticeable gains in recognition performance are realized when using the novel feature extraction approach. Finally, feature-fusion guidelines are proposed to provide the necessary insight to leverage the rich information effectively when fusing multiple biometric modalities at the feature level. These guidelines serve as the foundation to better understand and construct biometric systems that are effective in a variety of applications.
- Full Text:
- Date Issued: 2018
Information security concerns around enterprise bring your own device adoption in South African higher education institutions
- Authors: Sauls, Gershwin Ashton
- Date: 2016
- Language: English
- Type: Thesis , Masters , MSc
- Identifier: http://hdl.handle.net/10962/3619 , vital:20530
- Description: The research carried out in this thesis is an investigation into the information security concerns around the use of personally-owned mobile devices within South African universities. This concept, which is more commonly known as Bring Your Own Device or BYOD has raised many data loss concerns for organizational IT Departments across various industries worldwide. Universities as institutions are designed to facilitate research and learning and as such, have a strong culture toward the sharing of information which complicates management of these data loss concerns even further. As such, the objectives of the research were to determine the acceptance levels of BYOD within South African universities in relation to the perceived security risks. Thereafter, an investigation into which security practices, if any, that South African universities are using to minimize the information security concerns was carried out by means of a targeted online questionnaire. An extensive literature review was first carried out to evaluate the motivation for the research and to assess advantages of using Smartphone and Tablet PC’s for work related purposes. Thereafter, to determine security concerns, other surveys and related work was consulted to determine the relevant questions needed by the online questionnaire. The quantity of comprehensive academic studies concerning the security aspects of BYOD within organizations was very limited and because of this reason, the research took on a highly exploratory design. Finally, the research deliberated on the results of the online questionnaire and concluded with a strategy for the implementation of a mobile device security strategy for using personally-owned devices in a work-related environment.
- Full Text:
- Date Issued: 2016
- Authors: Sauls, Gershwin Ashton
- Date: 2016
- Language: English
- Type: Thesis , Masters , MSc
- Identifier: http://hdl.handle.net/10962/3619 , vital:20530
- Description: The research carried out in this thesis is an investigation into the information security concerns around the use of personally-owned mobile devices within South African universities. This concept, which is more commonly known as Bring Your Own Device or BYOD has raised many data loss concerns for organizational IT Departments across various industries worldwide. Universities as institutions are designed to facilitate research and learning and as such, have a strong culture toward the sharing of information which complicates management of these data loss concerns even further. As such, the objectives of the research were to determine the acceptance levels of BYOD within South African universities in relation to the perceived security risks. Thereafter, an investigation into which security practices, if any, that South African universities are using to minimize the information security concerns was carried out by means of a targeted online questionnaire. An extensive literature review was first carried out to evaluate the motivation for the research and to assess advantages of using Smartphone and Tablet PC’s for work related purposes. Thereafter, to determine security concerns, other surveys and related work was consulted to determine the relevant questions needed by the online questionnaire. The quantity of comprehensive academic studies concerning the security aspects of BYOD within organizations was very limited and because of this reason, the research took on a highly exploratory design. Finally, the research deliberated on the results of the online questionnaire and concluded with a strategy for the implementation of a mobile device security strategy for using personally-owned devices in a work-related environment.
- Full Text:
- Date Issued: 2016
Internal fingerprint extraction
- Authors: Darlow, Luke Nicholas
- Date: 2016
- Language: English
- Type: Thesis , Masters , MSc
- Identifier: http://hdl.handle.net/10962/2959 , vital:20347
- Description: Fingerprints are a non-invasive biometric that possess significant advantages. However, they are subject to surface erosion and damage; distortion upon scanning; and are vulnerable to fingerprint spoofing. The internal fingerprint exists as the undulations of the papillary junction - an intermediary layer of skin - and provides a solution to these disadvantages. Optical coherence tomography is used to capture the internal fingerprint. A depth profile of the papillary junction throughout the OCT scans is first constructed using fuzzy c-means clustering and a fine-tuning procedure. This information is then used to define localised regions over which to average pixels for the resultant internal fingerprint. When compared to a ground-truth internal fingerprint zone, the internal fingerprint zone detected automatically is within the measured bounds of human error. With a mean- squared-error of 21.3 and structural similarity of 96.4%, the internal fingerprint zone was successfully found and described. The extracted fingerprints exceed their surface counterparts with respect to orientation certainty and NFIQ scores (both of which are respected fingerprint quality assessment criteria). Internal to surface fingerprint correspondence and internal fingerprint cross correspondence were also measured. A larger scanned region is shown to be advantageous as internal fingerprints extracted from these scans have good surface correspondence (75% had at least one true match with a surface counterpart). It is also evidenced that internal fingerprints can constitute a fingerprint database. 96% of the internal fingerprints extracted had at least one corresponding match with another internal fingerprint. When compared to surface fingerprints cropped to match the internal fingerprints’ representative area and locality, the internal fingerprints outperformed these cropped surface counterparts. The internal fingerprint is an attractive biometric solution. This research develops a novel approach to extracting the internal fingerprint and is an asset to the further development of technologies surrounding fingerprint extraction from OCT scans. No earlier work has extracted or tested the internal fingerprint to the degree that this research has.
- Full Text:
- Date Issued: 2016
- Authors: Darlow, Luke Nicholas
- Date: 2016
- Language: English
- Type: Thesis , Masters , MSc
- Identifier: http://hdl.handle.net/10962/2959 , vital:20347
- Description: Fingerprints are a non-invasive biometric that possess significant advantages. However, they are subject to surface erosion and damage; distortion upon scanning; and are vulnerable to fingerprint spoofing. The internal fingerprint exists as the undulations of the papillary junction - an intermediary layer of skin - and provides a solution to these disadvantages. Optical coherence tomography is used to capture the internal fingerprint. A depth profile of the papillary junction throughout the OCT scans is first constructed using fuzzy c-means clustering and a fine-tuning procedure. This information is then used to define localised regions over which to average pixels for the resultant internal fingerprint. When compared to a ground-truth internal fingerprint zone, the internal fingerprint zone detected automatically is within the measured bounds of human error. With a mean- squared-error of 21.3 and structural similarity of 96.4%, the internal fingerprint zone was successfully found and described. The extracted fingerprints exceed their surface counterparts with respect to orientation certainty and NFIQ scores (both of which are respected fingerprint quality assessment criteria). Internal to surface fingerprint correspondence and internal fingerprint cross correspondence were also measured. A larger scanned region is shown to be advantageous as internal fingerprints extracted from these scans have good surface correspondence (75% had at least one true match with a surface counterpart). It is also evidenced that internal fingerprints can constitute a fingerprint database. 96% of the internal fingerprints extracted had at least one corresponding match with another internal fingerprint. When compared to surface fingerprints cropped to match the internal fingerprints’ representative area and locality, the internal fingerprints outperformed these cropped surface counterparts. The internal fingerprint is an attractive biometric solution. This research develops a novel approach to extracting the internal fingerprint and is an asset to the further development of technologies surrounding fingerprint extraction from OCT scans. No earlier work has extracted or tested the internal fingerprint to the degree that this research has.
- Full Text:
- Date Issued: 2016
- «
- ‹
- 1
- ›
- »