Computer Vision Lab
Dr Richard Green, Dr Mukundan.
Contact Dr. Richard Green.
The goal of computer vision is to compute properties of the three-dimensional world from images and video. Problems in this field include identifying objects and their motion. The computer vision lab supports the following research in this rapidly growing area of computing:
Orthogonal feature descriptors - Dr Mukundan
I had recently introduced a new class of orthogonal feature descriptors for applications in pattern recognition, image classification and object identification. This research has spawned further research in the areas of invariant descriptors using discrete orthogonal functions, image reconstruction and compression. Current studies include a comparative analysis of the performance of these functions and conventionally used functions for image compression (such as the DCT), and related compression algorithms.
Segmenting motion - Dr Richard Green
A hierarchical Hidden Markov Model (HMM) has been implemented to segment complex (human) motion. A new paradigm of dynemes uses an alphabet of movement units to support the HMM recognising hundreds of skills. This ongoing project is extending the model to support sub-domains of motion.
An expert’s skill overlays a training skill to also study accelerated sports skill acquisition. This research lays the foundation for more New Zealand Olympic medals by establishing New Zealand as a leader of accelerated skill acquisition research.
Tracking complex articulated motion - Dr Richard Green
A novel 3D colour body model is accurately sized and texture mapped to each person for more robust tracking using structure from motion. Tracking is further stabilised by estimating the joint angles for the next frame using a forward smoothing particle filter. Ongoing research examines optimising the particle filter and 3D model using single and multiple cameras. This research is also exploring the use of biomechanical overlays to accelerate movement skill acquisition in sports. This research enables coaches of all sports to enhance technique with a low-cost simple tool for biomechanical performance monitoring and analysis. This research is being used for gymnastics in Christchurch at Olympia Gymnastic Sports Inc: www.olympia.org.nz
Robot guidance and electric wheelchair collision avoidance - Dr Richard Green
Unlike many other approaches to date our investigation does not limit robot guidance to only either moving or turning. Our novel algorithm makes uses a Lucas Kanade method to calculate the optic flow over image pyramids in real-time. We are also applying this to collision avoidance for electric wheelchairs in collaboration with IRL and Dynamic Controls Ltd. This research enables the disabled to drive electric wheelchairs safely by preventing curb-side falls and collisions with narrow door-jams and to further improve their quality of life.
Robust tracking for cricket - Dr Richard Green
Using a state-based target candidates election algorithm, a finite state machine (FSM) is implemented to represent target object motion and elect target objects from candidates. This novel algorithm is robust since neither the target features nor its trajectory information is required to be retained properly. A cricket ball tracking/training game system has been implemented based on this algorithm. This research supports New Zealand remaining the leading cricket nation with real-time performance monitoring and game strategy support. It also supports children playing computer games while exercising and gaining movement skills.
Immersive simulation for snowboarding - Dr Richard Green
A Gyroboard www.gyroboard.com allows a snowboarder to improve their balance and co-ordination skills, strengthen their lower body, and train all year round, in an extremely enjoyable and challenging way. It can rock, tilt and spin to suit any level of ability. Imagine this hooked up to a gaming console, well Yuganeethen Yugaraja, Robert Lechte, and David Thomson did this by developing a new interface controller based on signals from a accelerometer. Scoring top points, the Gyroboard was found to be more Realistic and Enjoyable than an XBox 360 controller and the results from their project lead the way for the development for the ultimate electronic sports board.
Public interactive displays - Dr Richard Green
Research into robust interactions in public spaces with virtual characters. A Lord of the Rings interaction project uses an Orc to mimic movement by tracking users with a stereo camera. Public interaction with giant displays is novel research to lay the foundation for enhancing experience in public spaces to provide information and entertainment. This is being extended to immersive 3D interactions for games based research.
Biometrics - Dr Richard Green
Novel computer vision based gait and anthropometric biometrics have been recently developed and published. These biometrics can be further improved by being more robust with regard to loose clothing and carried objects. This poses interesting theoretical problems concerning segmenting loosely coupled moving objects. This research supports automatic non-invasive security monitoring.
Biomedical - Dr Richard Green
- Parkinson’s Disease - Progress has been made quantifying Parkinson’s Disease and other movement disorders in a clinical setting using computer vision gait analysis. This diagnostic tool enables clinicians to quantify and track treatment response going beyond the current subjective manual evaluations.
- Physiotherapy - The human-computer interaction focus within biomedicine is also monitoring movement repetitions in a physiotherapy environment. This overcomes the problem of unsupervised treatment by assisting the patient achieve the correct range of motion and repetitions with quantified history for the clinician.
- Wheelchair kinematics - Collaborative research with the Burwood Disability Centre, Dr Shane Gooch (from Mechanical Engineering) and Professor Alastair Rothwell (from Department of Orthopaedic Surgery and Musculoskeletal Medicine, Christchurch School of Medicine) to apply computer vision techniques to quantify post-op range-of-motion outcomes of a novel troid operation to tetraplegics to enhance their quality of life.
- Medical imaging - Low cost automatic quantifying of 3D volumes of brain lesions which is currently done manually in many research labs. This has already accelerated research data analysis in the Psychology Department.
For further information contact Dr Richard Green:
Department of Computer Science and Software Engineering (CSSE), University of Canterbury Christchurch New Zealand +64 3 3642987 x8436 - COSC, +64 21 331707 - cell