Showing items 1 - 14 of 14

Your selections:

A model based on computer vision for pose recognition in ballet

  • Authors: Fourie, Margaux
  • Subjects: Computer vision , Ballet , Pattern recognition systems
  • Language: English
  • Type: Masters (Thesis)
  • Identifier: http://hdl.handle.net/10210/417405 , uj:35348
  • Description: Abstract: The presence of computer vision technology is continually expanding into multiple application domains. With the increased availability of new camera and sensor technologies, it is possible to gather the necessary data required to apply computer vision in multiple environments. The task of recognising human activities using technology is one of the primary applications of computer vision. A human activity and an art form that is particularly attractive for the application of computer vision algorithms is ballet. Due to the well-codified poses, along with the challenges that exist within the ballet domain, automation for the ballet environment is a relevant research problem. This dissertation considers the use of computer vision methods to recognise different ballet poses. A literature review is done first to determine if it is a valid problem and explore the current methods used within ballet. Accordingly, the study proposes a model called BaReCo for ballet pose recognition using computer vision, which serves as a guide for the implementation of the prototype system. A specialised benchmark is then created to assess various facets of the model. The developed solution effectively makes use of a captured dataset, which was created for this study. The implemented computer vision pipelines contain various stages including pre-processing, localisation, feature extraction and classification. Results are consequently derived from the prototype to address the benchmark. The results of the benchmark for the BaReCo implementation, show that the study accomplishes the objective of recognising ballet poses using computer vision methods. Some key findings indicate that closely related poses are the potential cause for errors in recognition. The results also reveal that the use of novel deep learning techniques such as OpenPose and neural networks, along with traditional classification approaches, yield promising results. The study additionally provides a ballet pose dataset which serves as a contribution to the ballet and computer vision community. Future work for the study includes making improvements to the current implementation, as well as the application of other computer vision approaches on images and video data. The prototype system has validated the use of computer vision in the ballet domain to achieve pose recognition successfully. , M.Sc. (Computer Science)
  • Full Text:

Aircraft recognition using generalised variable-kernel similarity metric learning

  • Authors: Naudé, Johannes Jochemus
  • Date: 2014-12-01
  • Subjects: Image processing , Computer vision , Pattern recognition systems , Optical pattern recognition , Airplanes - Recognition
  • Type: Thesis
  • Identifier: http://ujcontent.uj.ac.za8080/10210/387397 , uj:13139 , http://hdl.handle.net/10210/13113
  • Description: M.Ing. , Nearest neighbour classifiers are well suited for use in practical pattern recognition applications for a number of reasons, including ease of implementation, rapid training, justifiable decisions and low computational load. However their generalisation performance is perceived to be inferior to that of more complex methods such as neural networks or support vector machines. Closer inspection shows however that the generalisation performance actually varies widely depending on the dataset used. On certain problems they outperform all other known classifiers while on others they fail dismally. In this thesis we allege that their sensitivity to the metric used is the reason for their mercurial performance. We also discuss some of the remedies for this problem that have been suggested in the past, most notably the variable-kernel similarity metric learning technique, and introduce our own extension to this technique. Finally these metric learning techniques are evaluated on an aircraft recognition task and critically compared.
  • Full Text:

An investigation into the parameters influencing neural network based facial recognition

  • Authors: Oerder, Stacy-Ann
  • Date: 2012-09-05
  • Subjects: Face perception - Data processing , Neural networks (Computer science) , Image processing - Digital techniques , Pattern recognition systems
  • Type: Thesis
  • Identifier: uj:9584 , http://hdl.handle.net/10210/7007
  • Description: D.Ing. , This thesis deals with an investigation into facial recognition and some variables that influence the performance of such a system. Firstly there is an investigation into the influence of image variability on the overall recognition performance of a system and secondly the performance and subsequent suitability of a neural network based system is tested. Both tests are carried out on two distinctly different databases, one more variable than the other. The results indicate that the greater the amount of variability the more negatively affected is the performance rating of a specific facial recognition system. The results further indicate the success with the implementation of a neural network system over a more conventional statistical system.
  • Full Text:

Asymmetrical three-phase fault evaluation in a distribution network using the genetic algorithm and the particle swarm optimisation

  • Authors: Shambare, Chikomborero
  • Date: 2020
  • Subjects: Genetic algorithms , Asymmetric synthesis , Short circuits , Electrical engineering , Pattern recognition systems
  • Language: English
  • Type: Masters (Thesis)
  • Identifier: http://hdl.handle.net/10210/413252 , uj:34808
  • Description: Abstract: Modern electric power systems are made up of three main sub-systems: generation; transmission; and distribution. The most common faults in distribution sub-systems are asymmetrical three-phase short circuit faults due to the fact that asymmetrical three-phase faults can be: line-to-line faults; two lines-to-earth faults; and single line-to-earth faults. This increases their probability of occurrence, unlike symmetrical three-phase faults which can only occur when all the three phases have been simultaneously shorted. Standard IEC 60909 and IEC 61363 provide all the basic information that is used for the detection of short circuit faults. However, the two standards use numerous estimates in their faults evaluation procedures. They estimate voltage factors (c), impedance correction factors (k), resistance to reactance ratios (R/X), resistance to impedance ratios (R/Z) and various other scaling factors for rotating machines. These IEC estimates are not evenly distributed throughout the 550kV and as such, they do not sufficiently cater for every nominal voltage. When the need arises, the user has to estimate these values accordingly. This research presents a genetic algorithm (GA) and a particle swarm optimisation (PSO) for the detection of asymmetrical three-phase short circuit faults within electric distribution networks of power systems with nominal voltages less than 550kV. GA and PSO are nature-inspired optimisation techniques. Although PSO has quick convergence, it suffers from partial optimism and premature stagnation. Some innovative coding adjustments were made in the creation of initial positions and particle distribution within the swarm. The GA struggles with: survival rates of individuals; stalling during optimisation; and proper gene replacements. Coding adjustments were also made to GA with regards to: strategic gene replacements; crossover when combining the properties of parents; and the arrangement of scores and expectation. Pattern search and Fmincon algorithms were also added to both algorithms as minimisation functions that commence after the evolutionary algorithms (EAs) terminate. The EAs were initially tested on the Rastrigin and Rosenbrock functions to ensure their efficiencies. During fault detection, the developed EAs were used to stochastically determine some of the most crucial estimates (R/X and R/Z ratios). The proposed methodology would compute these values on a case-to-case basis for every optimisation case with regards to the parameters and unique specifications of the power system. The EAs were tested on a nominal voltage that is properly catered for by Standard IEC. They obtained ratios, impedances and currents that were within an approximate range to the IEC values for that nominal voltage. This further implies that EAs can be reliably used to: stochastically determine these ratios; compute impedances; and detect fault currents for all the nominal voltages including those that are not sufficiently catered for by Standard IEC. Since R/X and R/Z ratios play a key role in determining the upstream and fault point impedances, the proposed methodology can be used to compute much more precise fault magnitudes at various network levels thereby setting up and repairing power systems sufficiently. , M.Ing. (Electrical and Electronic Engineering Science)
  • Full Text:

Computer-aided detection of tuberculosis in chest radiographs

Image classification using machine learning techniques

  • Authors: Nkonyana, Thembinkosi Nelson
  • Date: 2016
  • Subjects: Image processing - Digital techniques , Image analysis , Remote-sensing images , Pattern recognition systems , Machine learning
  • Language: English
  • Type: Masters (Thesis)
  • Identifier: http://hdl.handle.net/10210/212965 , uj:21061
  • Description: Abstract: Image classification entails the important part of digital image and has been very essential in the application of remote sensing systems, thus the demand for research to find advanced algorithms and tools to solve problems experienced in classification has shown great increase in interest over the years. In this day and age, remote sensing has globally being applied with the use of current advanced satellite systems and sensors, but the need to provide analysis and decision making has been a challenge. The contribution of this dissertation is an empirical comparison (evaluation) of five machine learning (ML) techniques, in terms of classifying satellite images. The ML techniques consist of Random Forest (RF), K Nearest Neighbour (k-NN), Naïve Bayes (NB), Multi-layer Perceptron (MLP) and Support Vector Machines (SVM). The evaluation of these five techniques is based on a selection of six performance measures, such as [Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Relative Absolute Error (RAE), Root Relative Squared Error (RRSE), Precision, and Receiver Operating Characteristics (ROC)] Three publicly available remote sensing datasets are utilised for this task. The experimental results show that RF achieved higher accuracy rates, with robust performance, followed by MLP, k-NN, NB and SVM classifier exhibiting the worst performance. Hence, the use of ML for image analysis and pattern recognition is a promising approach. , M.Phil. (Electrical Engineering)
  • Full Text:

Intelligent system for automated components recognition and handling

  • Authors: Findlay, Peter
  • Date: 2012-02-06
  • Subjects: Computer vision , Artificial intelligence , Pattern recognition systems
  • Type: Thesis
  • Identifier: uj:2012 , http://hdl.handle.net/10210/4365
  • Description: M.Ing. , A machine vision system must, by definition, be intelligent, adaptable and reliable to satisfY the objectives of a system that is highly interactive with its dynamic environment and therefore prone to outside error factors. A machine vision system is described that utilizes a 2D captured web cam image for the purpose of intelligent object recognition, gripping and handling. The system is designed to be generic in its application and adaptable to various gripper configurations and handling configurations. This is achieved by using highly adaptable and intelligent recognition algorithms the gathers as much information as possible from a 2D colour web cam image. Numerous error-checking abilities are also built into the system to account for possible anomalies in the working environment. The entire system is designed around four separate but tightly integrated systems, namely the Recognition, Gripping and Handling structures and the Component Database which acts as the backbone of the system. The Recognition system provides all the input data that is then used for the Gripping and Handling systems. This integrated system functions as a single unit but a hierarchical structure has been used so that each of the systems can function as a stand-alone unit. The recognition system is generic in its ability to provide information such as recognized object identification, position and other orientation information that could be used by another handling system or gripper configuration. The Gripping system is based on a single custom designed gripper that provides basic gripping functionality. It is powered by a single motor and is highly functional with respect to the large range of object sizes that it can grip. The Handling Sub-system controls gripper positioning and motion. The Handling System incorporates control of the robot and the execution of both predetermined and online adaptable handling algorithms based on component data. It receives data from the Component database. The database allows the transparent ability to add and remove objects for recognition as well as other basic abilities. Experimental verification of the system is performed using a fully integrated and automated program and hardware control system developed for this purpose. The integration of the proposed system into a flexible and reconfigurable manufacturing system is explained.
  • Full Text:

Mechanical condition monitoring of impulsively loaded equipment using neural networks

Multimodal verification of identity for a realistic access control application

  • Authors: Denys, Nele
  • Date: 2008-11-18T09:08:58Z
  • Subjects: Pattern recognition systems , Human face recognition (Computer science) , Optical character recognition devices , Image processing , Automatic control
  • Type: Thesis
  • Identifier: uj:14730 , http://hdl.handle.net/10210/1734
  • Description: D. Ing. , This thesis describes a real world application in the field of pattern recognition. License plate recognition and face recognition algorithms are combined to implement automated access control at the gates of RAU campus. One image of the license plate and three images of the driver’s face are enough to check if the person driving a particular car into campus is the same as the person driving this car out. The license plate recognition module is based on learning vector quantization and performs well enough to be used in a realistic environment. The face recognition module is based on the Bayes rule and while performing satisfactory, extensive research is still necessary before this system can be implemented in real life. The main reasons for failure of the system were identified as the variable lighting and insufficient landmarks for effective warping.
  • Full Text:

Network intrusion detection system using neural networks approach in networked biometrics system

  • Authors: Mgabile, Tinny
  • Date: 2014-04-09
  • Subjects: Computer networks - Security measures , Neural networks (Computer science) , Pattern recognition systems , Biometric identification
  • Type: Thesis
  • Identifier: uj:10528 , http://hdl.handle.net/10210/10054
  • Description: M.Phil. (Electrical and Electronic Engineering) , Network security has become increasingly important as more and more applica- tions are making their way into the market. The research community has proposed various methods to build a reliable network intrusion detection system to detect unauthorised activities in networked systems. However many network intrusion detection systems that have been reported in literature su er from an excessive number of false positives, false negatives, and are unable to cope with new, elegant and structured attacks. This is mainly because most network intrusion detection systems rely on security experts to analyze the network tra c data and manually construct intrusion detection rules. This study proposes to use a machine learning technique such as neural network approach to anomaly based network intrusion detection system (NIDS). The main objective for this study is to construct an NIDS model that will produce approx- imate to zero false positive or no false positive at all and have high degree of accuracy in detecting network attacks. The neural network (NN) model is trained on a biometric networked system dataset simulated in the study, containing strictly replayed and normal network tra c that encourage the development of the pro- posed NIDS. By analyzing the NN{based NIDS results, the study reached the false positive rate of 0, and high accuracy rate of 100 percent. To support the results obtained in this study, the performance of the NN{based NIDS was compared to two other classi cation methods (k{nearest neighbor algorithm (KNN) and Naive Bayes). The results obtained from KNN and naive Bayes were 99.87 and 99.75 percent respectively. These results show that the proposed model can successfully be used as an e ective tool for solving complicated classi cation problems such as NIDS.
  • Full Text:

Reconnaissance and assessment of iris features towards human iris classification

  • Authors: Mabuza-Hocquet, Gugulethu P.
  • Date: 2018
  • Subjects: Biometric identification , Optical pattern recognition , Pattern recognition systems
  • Language: English
  • Type: Doctoral (Thesis)
  • Identifier: http://ujcontent.uj.ac.za8080/10210/368077 , http://hdl.handle.net/10210/286038 , uj:30944
  • Description: Abstract: The use of counterfeit documents, human trafficking, identity theft, terrorist attacks and cyber-attacks are amongst many of the challenging problems and threats that the whole world is exposed to. Today’s era of advancing technologies means that more people and more devices are connected, and can easily communicate to and with each other via the Internet. Although global communication is a necessity, it also entails the risk of exposing personal or highly classified data to any sort of malicious exploitation. The need to increase security has seen the use of biometrics as a necessary alternative. The growing role of biometric methods have resulted to countries such as India, the United Emirates, Japan, etc. to implement biometric systems for applications in national ID cards, border security, immigration control, and law enforcement, retail stores, banks and government facilities. Amongst the various biometric modalities, the human iris is regarded as the most accurate, and as such has drawn a lot of attention and gained momentum for over a decade due to the uniqueness, reliability and stability of iris features over a person’s lifetime, as well as the high accuracy achieved for authentication, and ease of image acquisition. A typical iris recognition system (IRS) consists of four modules namely iris segmentation, normalisation, feature extraction and template matching. Each module has automated traditional algorithms that have been successfully used solely for the purpose of uniquely identifying and verifying a person within a large database of enrolled individuals. The drawback of the classical iris segmentation algorithm for instance, is that is assumes that the pupil and iris boundaries are concentric circles, that is, they share the same center, which is not generally the case. The normalisation stage uses the rubber sheet model to transform the segmented iris from a Cartesian plane to polar coordinates to cater for... , D.Phil. (Electrical and Electronic Engineering)
  • Full Text:

Super features: a probabilistic approach to feature matching and correction

SWORM : a Semantic Web Object Recognition Model

  • Authors: Minnaar, Ursula
  • Date: 2011-10-11T06:59:58Z
  • Subjects: Semantic Web , Object-oriented methods (Computer science) , Pattern recognition systems , SWORM
  • Type: Thesis
  • Identifier: uj:7231 , http://hdl.handle.net/10210/3886
  • Description: D.Phil. , The Semantic Web is an extension of the current Web. The goal of the Semantic Web is to give information “well-defined meaning, enabling computers and people to work in better cooperation” (Berners-Lee, Hendler, & Lassila, 2001). While the Semantic Web is not artificial intelligence, it does involve defining information in such a way that it can be more easily “understood” by machines. The Semantic Web builds upon the advantages offered by XML, and introduces languages such as the Resource Description Framework to address some of the shortcomings of XML. It uses ontologies to provide a mechanism for information processing on the Web. Object recognition involves the recognition of unknown objects and is usually divided into two types of recognition: object classification and object identification. Classification refers to the categorization of an unknown object into a known group, while identification is the matching of an unknown object against the memory of a known object. Most object recognition techniques, regardless of the recognition type, involve the extraction of some type of processable data from objects, and the subsequent comparison of the extracted information. The research presented in this thesis investigates the possibility of using the languages developed for the Semantic Web to perform some type of object recognition. It is hoped that by treating object recognition as an information management task, the advantages provided by the information-centric Semantic Web can be used in good stead. The goal of the research is to determine whether ontology-based descriptions can be created, whether such descriptions can be compared, and to what extent the use of the Semantic Web could enhance information sharing in object recognition. In order to investigate these questions, the research defines the Semantic Web Object Recognition Model. The model provides a recognition framework that uses ontologies to create and compare object descriptions. The model also suggests the use of web agents to perform distributed object comparisons across the relevant domain.
  • Full Text:

VFFD : voting features for face detection

  • 1
  • Disclaimer
  • Privacy
  • Copyright
  • Contact