Student projects

I have an interest in any potential applications of Image Processing/ Computer Vision to solve/advance real-life problems. Below a few applied projects in hot topics are detailed. I have other interesting projects both theoretical and practical. Visit me (room 391) or email me pDOTdelmasATaucklandDOTacDOTnz


Some of these projects may look hard to deal with but they can be split in different sub-projects and/or students depending on the project length (from summer project to PhD studies).

2/3D Face texture analysis and synthesis (ongoing)

In collaboration with A. Professor Georgy Gimel'farb

In our way to a realistic synthetic talking face, texture analysis is one of the main key for the future of this field. So far, texture rendering in synthetic face is usually too regular and did not render the exceptional complexity of human face texture. Your goal will be to find the characteristics of the face texture, extract them and be able to synthesize it as close as possible to the original ones.

Aim

Develop programs able to cope with texture analysis, extraction and synthesis from 2D face image to 3D face model.

Gain

Image filters, 3D face synthesis

3D Face tracking and synthesis (ongoing)

More and more applications (games, human-computer interface, real virtuality)  are looking for realistic synthetic talking faces. If face synthesis is now on the good way with human-like behavior for the last up-to-date creations (Final Fantasy), there are still problems to link 2D face tracking to 3D synthesis.
You will have to track faces, extract face features (nostrils, eyebrows, eyes, mouth) and recreate a realistic dynamic 3D face. You will use SONY webcams available at the Computer Vision Lab. Systems allowing 3D static face synthesis are already available in the visualization lab.

Aim

Create and track 3D realistic human face from 2D camera views.

Gain

Stereo vision, 3D face animation

Deformable Models for face features extraction

This is more an exploratory project on how image processing methods categorized as deformable models have been used to achieve lip and face features extraction. The goal could be to define a suitable strategy for real-time reliable face and lip tracking. This will require high level of understanding in mathematics and excellent programming skills. In return, you will have an expert comprehension of these methods widely used in Medical Imaging, Virtual reality and Human-Computer Interfaces. This project could be extend far beyond face and engulfed human.

Aim

A complete 2D to 3D hyper-realistic human face for Internet applications

Gain

A lot

2D-3D Face recognition

In collaboration with Georgy Gimel’farb

This project aims at recognizing people for tasks such as restricted area control or entrance statistics.
The system should be able to recognize people by extracting face features and comparing these parameters to a database you will have to build. Several systems are already exiting but they generally rely on pre-scaled images and constant lighting conditions while using more general face features such as eigenfaces. You will use SONY web cam available in the Computer Vision Lab and 3D acquisition system to create your 3D face database.

Aim

Automatically recognize (and grant access to) persons entering the CVlab

Gain

Database management. High level Image processing operators.

Face expressions recognition and/or gesture recognition for user interface and robots control

This project aims at recognizing face expressions and/or arms and/or hands signs to control robots and interface functionalities.
One idea is to control robots through conventional signs (previously categorized). These signs could be coming either from face expressions (7 universally classified expressions from anger to surprise) or hands (fist closed, finger pointing a direction) and arms (angles between forearms and upper arms could indicate angles between joints of a robot manipulator, hand claws open or closed could mean robot gripper open or closed)  3D position. Here, the idea is to develop a real-time system able to send a few commands (direction, speed, motion) to robots. The robotics part (robots positioning and control) of the project is already fully working on a realtime basis in the Vision lab (city Campus). This project is mainly vision-oriented with a Computer graphic flavour.
 

Aim

Automatically recognize signs to control robots.

Gain

Higher level image operators, 3D positioning, robots control.
 

Video-surveillance of public areas.

This project aims at recognizing and tracking human characters passing or evolving through the field of the camera. The idea is to isolate people with unconventional behaviors. Possible applications could be automatic parking surveillance, public hall surveillance. Such approaches have already been implemented for soccer players tracking using HMM methods.

Aim

Automatically detect and track humans evolving in the field of the camera. Classify pedestrians behaviors and focus on suspicious ones.

Gain

Pattern recognition, Kalman filtering, Classification, Camera synchronization.
 

3D movies

Using 3D acquisitioned vices hosted in the vision lab you will be able to acquire 3D images and short sequences (face, hand,.. ). Your work will be to develop an easy way to visualise these movies, extend along temporal dimension existing stereo algorithms to explore the abilities of the acquisition system to be exported outside the lab and determine whether the frame rate may be increased to almost real-time in any condition.

Aim

Transform the 3D image acquisition system into a 3D robust movie acquisition system

Gain

Image processing, coding, 3D technology, stereo matching

Image Processing for soil sciences applications

In collaboration with Georgy Gimel’farb, Karin Mueller (HortResearch, Hamilton) Jorge Marquez (UNAM, Mexico) and Celine Duwig (IRD, France)

The Earth’s skin, the soil, is the host of the world’s most diverse microbial communities that exist in various refugia sheltered by the soil’s heterogeneous structure and diverse constituents. The inherent heterogeneity of this natural porous medium also significantly affects water flow and reactive solute transport, which impact on these habitats. To advance the understanding of the functioning of the Earth’s skin and to generate new knowledge about reactive solute flow in heterogeneous media, imaging and image processing techniques can be of great help. Currently the project has completed the 2D structure analysis of the soils studied and we are progressing towards 3D imaging and ensuing analysis of soil structure. Several subjects may be offered with possible internships in France or Mexico.

Aim

Design or use imaging apparatus delivering data as requested by soil scientists.

Extract meaningful soil properties from 2D, 3D and 4D datasets.

Visualise and model the soil properties and behaviour.

Gain

Image processing, coding, 3D technology, stereo matching

 

Feel free to visit me (room 391, CS City campus) or Georgy (room 390, CS City campus) if you have any ideas or projects that might interest me.
 

More soon

Back