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:
- 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:
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:
- 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:
- «
- ‹
- 1
- ›
- »