Visual Processing for the Bionic Eye
Research and development of visual processing for low vision devices and the bionic eye
In 2004 there were 50,000 people legally blind in Australia, with numbers expected to increase to 87,000 by 2024 with the ageing population ("Eye Research Australia Clear Insight: The Economic Impact and Cost of Vision Loss in Australia" by Centre for Eye Research Australia).
Although there are many useful devices on the market to assist individuals with vision impairment, there is a lack of the type of sensor-based assistant systems that are newly appearing in cars (.e.g, lane departure, collision warning, navigation).
This project aims to develop a new generation of assistive devices (stand-alone wearable devices that individuals can use without medical intervention) based on computer vision processing: it will produce prototype devices that aim to demonstrate effective assistance for individuals with vision impairment.
VIBE is also contributing expertise in computer vision processing to Bionic Vision Australia which is funded by the Australian Research Council. The Bionic Vision Australia partnership aims to build the first Australian bionic eye implant whereby individuals may recover some of the lost vision via electrical stimulation of the retina. Vision processing will be one of the key components of a bionic eye as it will enable efficient encoding of high resolution images into a set of stimulation signals on a retinal implant.
Research Outcomes
Those may be described as improving/restoring
a) the ambulatory navigation (e.g. obstacle avoidance)
b) the interaction with the environment (e.g. face detection and recognition, reading of text/symbols, etc)
Who will benefit?
Individuals who have little or no vision as a consequence of diseases like Retinitis Pigmentosa (RP) and age-related macular degeneration (AMD).
Key Features
- Vision Processing
- computation of stable depth maps, free space encoding for ambulatory navigation
- face detection and recognition and general object detection and recognition
- Visual Neuroscience
- building a retinal model for simulation of the visual function
- building a visual cortex model for simulation of perception
Results to Date
We have developed new camera ego-motion estimation algorithms by using a multi-camera rig. The new algorithms are able to accurately and reliably compute the moving camera rig's instantaneous position and orientation though the camera observing its dynamic environment.
Further, we have developed efficient face detection algorithms which outperform the state-of-the-art (see paper by Shen, Welsh and Wang below). We are currently working on robust face recognition.
Research Team
Dr Nick Barnes (project leader), Viorica Botea, Professor Richard Hartley, Dr Xuming He, Dr Hongdong Li, Yi Li, Dr Paulette Lieby, Dr Nianjun Liu, Dr Chris McCarthy, Dr Abd-Krim Seghouane, Dr Chunhua Shen.
PhD students: Kharum Aftab, Junae Kim, Hanxi Li, Samunda Parera, Kyoungup Park.
Expressions of Interest
If you consider doing your PhD with us, contact Dr Paulette Lieby. Information about the NICTA enhanced PhD program may be found here; and about the candidature process here.
Publications (subset)
John Lim, Nick Barnes and Hongdong Li, Estimating Relative Camera Motion from the Antipodal-Epipolar Constraint, in IEEE Trans PAMI, Accepted April 2010.
Chunhua Shen and Hanxi Li, On the dual formulation of boosting algorithms, in IEEE Trans PAMI, Accepted Dec, 2009.
Jae-hak Kim, Hongdong Li, and Richard Hartley, Motion Estimation for Non-Overlapping Multi-Camera Rigs: Linear Algebraic and L-infinity Geometric Solutions, IEEE Trans on PAMI, pp. 1-8, Nov, 2009.
Rene Vidal and Richard Hartley, Three-View Multibody Structure from Motion, in IEEE Trans PAMI, 30(2), 214-227, 2008.
Chunhua Shen and Sakrapee Paisitkriangkrai and Jian Zhang, Efficiently Learning a Detection Cascade with Sparse Eigenvectors, in IEEE Trans Image Processing, Accepted June 2010.
Peter Carr and Richard Hartley, Minimizing Energy Functions on 4-connected Lattices using Elimination, IEEE International Conference on Computer Vision (ICCV09), Kyoto, Japan, 2009.
Chunhua Shen, Junae Kim, Lei Wang, and Anton van den Hengel, Positive semidefinite metric learning with Boosting, in NIPS, 2009.
Chunhua Shen, Alan Welsh, and Lei Wang, PSDBoost: Matrix-generation linear programming for positive semidefinite matrices learning, in Advances in Neural Information Processing Systems (NIPS'08), pages 1473-1480, December 2008. Vancouver, B.C., Canada. MIT Press.
John Lim, and Nick Barnes, Robust Visual Homing with Landmark Angles, In Proc Robotics Science and Systems (RSS), June 2009, Seattle, USA.
Qinfeng Shi and Hanxi Li and Chunhua Shen, Rapid face recognition using hashing, IEEE Conference on Computer Vision and Pattern Recognition (CVPR'10), San Francisco, USA, June 2010.Hongdong Li, Efficient Reduction of L-infinity geometry problems, in proc IEEE CVPR 2009, June, Miami, USA.
Sakrapee Paisitkriangkrai, Chunhua Shen, Efficiently Training a Better Visual Detector with Sparse Eigenvectors, In: IEEE Conference on Computer Vision and Pattern Recognition (CVPR'09), June, Miami, USA.
John Lim and Nick Barnes, Directions of egomotion from antipodal points, in IEEE Conference on Computer Vision and Pattern Recognition (CVPR'08), Anchorage, USA.