Undergraduate Thesis Project Database: Search Results (9 records found)

NumberTitle
FCN013-D object reconstruction using structured lighting
FCN02Vibratory feeder analysis and control
FCN03Robot soccer vision system
FCN04Froth height/texture measurement and lighting investigation
FCN05Sensor and actuator hardware platform
FCN06Hyperspectral imagery using an imaging spectroscope
FCN073-D diamond shape classification and value estimation
FCN08Motion estimation for particle handling
FCN09Detecting TB in microscope images

ID:  FCN01
Title:  3-D object reconstruction using structured lighting
Description:  By projecting a light grid onto a 3-D object and observing the curvature of the lines with a camera, the shape of the object can be reconstructed. The aim of this project is to develop a general-purpose reconstruction system using a video camera and a standard digital video projector as a light source.

Using a video projector as a light source gives one a high degree of flexibility, and raises the possibility of changing the light pattern over time to resolve ambiguities and other difficult cases. This project therefore has considerable scope for imaginative investigation.
Deliverables:  A working prototype and a set of techniques for doing the reconstruction.
Skills/requirements:  A good understanding of geometry, particularly in 3-D. Ability to contend with fairly advanced mathematical problem formulations.
Extra:  
Area:  Image and video processing, computer vision.

ID:  FCN02
Title:  Vibratory feeder analysis and control
Description:  We are working together with Debeers on building a diamond-sizing machine, which uses multiple cameras and computer vision techniques to obtain particle size estimates. For correct operation the diamonds must be dropped through one at a time, at rates of up to 20 per second. Currently it is envisaged that diamonds will be fed into the system using a vibratory feeder. The feeder takes the form of a V-shaped channel that holds the diamonds in a long line. Vibration causes the particles to migrate down this channel, until they fall off the end and enter the machine.

The camera hardware requires that diamonds enter the system at fairly regular time intervals. For this reason we would like to control the feed-rate by varying the vibration amplitude. Actual feed rates can be measured either using photodetectors, a light curtain, or a peripheral camera system. The aim of this project is to investigate and analyse methods for controlling the particle feed into the system, and come up with a configuration and implementation that meets the requirements. We will have a vibratory feeder on loan to us for the duration of the project.
Deliverables:  A working prototype and a control strategy
Skills/requirements:  Knowledge of control theory, ability/aptitute to interface signals to digital (probably PC) hardware
Extra:  
Area:  Control and instrumentation, possibly image processing

ID:  FCN03
Title:  Robot soccer vision system
Description:  UCT's robot soccer team is reaching maturity. However, a fast and robust vision system is required to provide information regarding the positions of the robot players and the ball on the field. Last year we developed a rudimentary algorithm that should meet the requirements - what remains is to bed down the implementation, optimise the speed, and minimise the error rates in the reported positions.

The system is computationally demanding - position estimates for each of the 10 robots on the field are required at least 15 times per second (and preferably 30 times per second). To achieve this type of performance on a PC-based system requires using the SSE instruction set on the Pentium processor. These are a low-level set of instructions that facilitate parallel and highly pipelined computation on large quantities of data. Fortunately, Intel have released a moderately high-level set of C libraries that make using this functionality quite easy.

We would also like to consider the possibility of using a separate high frame- rate camera to locate the position of the ball.
Deliverables:  A working vision system, and an evaluation of its performance
Skills/requirements:  A good working knowledge of C and Matlab, and an understanding of fairly low-level computer hardware and peripherals. This project is not suitable for someone with only moderate programming skills.
Extra:  Intel has SSE performance libraries for image and video processing, signal processing, and basic math operations. If ever you plan to implement high- performance algorithms, there is a huge amount of value in becoming conversant with these libraries.
Area:  Basic image processing and pattern recognition, networking, high-performance computing

ID:  FCN04
Title:  Froth height/texture measurement and lighting investigation
Description:  The mining industry makes extensive use of froth flotation for mineral extraction. Currently the only way to control the flotation process involves visually inspecting the froth, and obtaining measurements which are subsequently used for control purposes.

UCT has developed an industrial package for making certain froth measurements using a camera. However, some measurements are difficult to make using the current system. We would therefore like to investigate alternative measurements that can be obtained by illuminating the froth with a laser line, and measuring the resulting light pattern with a camera. In principle such structured lighting can provide froth height and texture information. The extent of the scattering may also indicate froth composition.

Many mineral processing plants are being built without roofs, which introduces a problem of shadows and sunlight on the froth surface. We would therefore also like to investigate the use of narrowband light sources and matching camera filters to minimise this effect.
Deliverables:  
Skills/requirements:  Some appreciation for 3-D geometry
Extra:  
Area:  Image processing, computer vision, elementary physics

ID:  FCN05
Title:  Sensor and actuator hardware platform
Description:  PCs are excellent computing platforms, even for high-performance industrial applications. However, their capability for dealing with accurate timing and synchronisation is quite limited. These tasks are commonly assigned to a microprocessor or microcontroller subsystem, which can communicate with the host but essentially acts independently.

The aim of this project is to design and test a microprocessor subsystem for dealing with sensors and actuators required in industrial machine vision systems. Typical sensors include rotary encoders and various types of photodetectors, while the actuators are needed to control the switching of lights and ejectors. Some measure of feedback control functionality is also often required to keep various motor and conveyor components synchronised.

The system will probably take the form of a high-end PIC microprocessor, and we would like to have USB connectivity to the host PC.
Deliverables:  A hardware platform designed according to the requirements, and software routines for performing common IO tasks.
Skills/requirements:  A reasonable understanding of microprocessor and computer systems
Extra:  
Area:  Hardware and software for embedded systems

ID:  FCN06
Title:  Hyperspectral imagery using an imaging spectroscope
Description:  We will soon be acquiring an imaging spectroscope from the Finnish company SpecIm (www.specim.fi). This is a very interesting piece of equipment - instead of just measuring red, green, and blue components of the light colour at a camera pixel, it measures the complete spectrum of the light over all optical wavelengths. There is considerably more information in this complete spectrum - two colours we perceive as the same shade of green, for example, may appear very different under a spectroscope.

We would like to connect this device to a CMOS camera, and use it for both industrial and scientific purposes. The intention of this project is therefore to explore different ways of using the spectroscope in these types of applications. One specific problem involves calibrating the system to permit quantitative measurements to be made. Other applications involve mineral identification and classification, which has relevance to some projects we are currently running with Debeers and Anglo Platinum.
Deliverables:  Methods and ideas for using the spectroscope for typical applications
Skills/requirements:  Some knowledge and appreciation of physics and physical principles
Extra:  To generate a complete image from this imaging spectroscope requires pushbroom scanning. A computer-controlled conveyor belt would be useful for general application, and we may choose to construct one during the project
Area:  Physics, imaging, image processing

ID:  FCN07
Title:  3-D diamond shape classification and value estimation
Description:  We are working together with Debeers on building a diamond-sizing machine, which uses multiple cameras and computer vision techniques to obtain particle size estimates. However, the solution we have developed provides far more than just a volume estimate - accurate 3-D representations of the diamonds are generated as part of the procedure.

The aim of this project is to develop methods of classifying and identifying particular diamonds from this 3-D information. This has a number of applications, ranging from estimating diamond value to detecting diamond theft. The project will involve coming up with features for describing 3-D shape, and applying fairly classical pattern recognition methods to these features.
Deliverables:  A specified set of features and estimates of the classification performance
Skills/requirements:  Aptitude for 3-D geometry, some probability theory
Extra:  
Area:  Computer vision, pattern recognition

ID:  FCN08
Title:  Motion estimation for particle handling
Description:  Diamond breakage is a big problem in the diamond mining industry. Debeers are therefore involved in a set of projects to formulate diamond breakage models, and validate them against actual data. In many cases these models relate to high-grade streams of liberated diamonds.

One of the diamond breakage models currently under investigation takes as parameters quantities such as velocity and density of particles. We have available some very high speed video footage (~250fps) of diamonds in free-fall after coming off of a conveyor. The aim of this project is to analyse this data and obtain motion estimates for diamonds at different locations in the particle stream. Some very preliminary investigation indicates that block motion and optical flow techniques may be suitable.
Deliverables:  A motion estimation algorithm, and an evaluation of its performance.
Skills/requirements:  Matlab will be required, and knowledge of C may be desirable.
Extra:  
Area:  Image and video processing

ID:  FCN09
Title:  Detecting TB in microscope images
Description:  The Department of Human Biology at medical school have constructed a microscope which can automatically acquire colour images of TB sputum samples. The aim of this project is to use image processing and pattern recognition techniques to automatically identify TB bacilli in the images.
Deliverables:  A working algorithm
Skills/requirements:  An interest in medical imaging
Extra:  
Area:  Image processing, pattern recognition