research
U.S. funds research towards street intersection analyzer for the blind
The U.S. National Institutes of Health (NIH) is providing $400,000 in funding to Dr. James Coughlan, of the Smith-Kettlewell Eye Research Institute, to develop tools to help the visually impaired. Building on the team's existing computer vision software for detecting crosswalks and 'walk' lights, the new system will help the blind and visually impaired navigate across intersections via continous feedback. The goal is to deploy the system in the form of free smartphone apps for common platforms such as Android and iOS so that users will not have to carry a separate device.
Tracking all the players on the field
A team of researchers from Switzerland's Ecole Polytechnique Fédérale de Lausanne (EPFL) have developed a system that can continously track every player on the field in a sporting event, even through occlusion events. The system is entirely based on eight cameras positioned around the field, and does not require the players to wear any devices. The research was presented at ICCV 2011 this week. More information is available in a press release.
Students win competition with Kinect visual gait analysis
In the recent Siemens Competition in Math, Science and Technology held at Georgia Tech, high school seniors Ziyuan Liu and Cassee Cain took first place in the regional championship. Their project used the Xbox Kinect and computer vision to analyze human gait. The Oak Ridge High School seniors will advance to the national competition on Dec. 2-5 in Washington D.C., where they will compete for a $100,000 scholarship. More information is available in a Knoxnews article.
New MIT algorithm speeds up MRI scans
Researchers from MIT (Elfar Adalsteinsson and Vivek Goyal) have developed a new imaging algorithm that could reduce magnetic resonance imaging (MRI) scans from 45 to 15 minutes. Using information from the MRI's first contrast scan, the algorithm predicts the likely object boundaries between different types of tissues in subsequent scans. The researchers are now working to improve the algorithm and accelerate the processing time on a GPU. More information is available in a MIT news article.
Stanford Decaptcha identifies CAPTCHA weaknesses
Researchers at Stanford University have developed a tool to automatically decipher and defeat CAPTCHAs (Completely Automated Public Turing tests to tell Computers and Humans Apart). The research team (Elie Bursztein, Matthieu Martin, and John C. Mitchel) developed "Decaptcha" using various methods of cleaning up intentional background noise and breaking text strings into individual characters for easier recognition. They ran against CAPTCHAs used by 15 high-profile Web sites, and the only tested site that could not be broken was Google. Several recommendations to improve CAPTCHA security were provided, which include randomizing the length of the text string, randomizing the character size, applying a wave-like effect to the output, and using a collapsing background with lines. More information is available in a Network World article.
Related articles:
iSpy: long-distance shoulder-surfing
New Scientist reports on research on automated shoulder-surfing presented at the Conference on Computer and Communications Security in Chicago. A University of North Carolina at Chapel Hill research team, led by Dr. Jan-Michael Frahm, developed computer vision software that detects the visual feedback signals provided by touchscreen keyboards; e.g., the momentary magnification of each letter as it is typed. The news article suggests that it would be possible to to read the words typed by a user from 60 meters away, using HD video captured by an SLR camera, and from 3 meters using a cell phone camera. Click here for the paper.
Combining knowledge with data for generalizable and robust visual learning
The U.S. National Science Foundation (NSF) has awarded a $200,000 grant to Rensselaer Polytechnic Institute (RPI) to study the combination of heterogeneous domain knowledge with data-driven machine learning algorithms for computer vision problems. The researchers, led by Dr. Qiang Ji, will systematically identify knowledge from different sources and develop methods of integrating it with data-driven algorithms.
Research abstract:
Computer vision has made tremendous progress in the past decades, partially enabled by the advanced machine learning techniques. But compared with human perception, computer vision remains primitive. One contributing factor for this is the data-driven nature of the current learning algorithms and their inability to incorporate any related knowledge. The data-driven methods tend to be database-specific and cannot generalize well to unseen data. This project addresses this issue through the introduction of a knowledge-augmented statistical learning framework. Within this framework, knowledge and data can be systematically exploited, captured, and are principally integrated to jointly train a vision algorithm. Developing such a framework, however, is challenging since the domain knowledge often exists in different and diverse formats, typically inaccessible to the data-driven statistical machine learning methods. To overcome this challenge, the research team systematically converts domain knowledge into either the constraints on the model or into pseudo-data, whereby they can be incorporated into the statistical learning methods. The project includes systematic identification of knowledge from different sources and concrete mechanisms to capture the knowledge and to convert them into formats easily accessible to the automatic machine learning methods. The project also involves demonstrating the effectiveness of the proposed framework for certain computer vision problems.
Computer vision makes progress on solving CAPTCHAs
Dr. Jeff Yan's team at Newcastle University, U.K., has developed computer vision algorithms for solving CAPTCHAs. CAPTCHAs -- short for Completely Automated Public Turing Test to Tell Computers and Humans Apart -- are used by many websites to ensure that users signing up for accounts or posting comments are real people and not automated programs. CAPTCHAs work by displaying letters and numbers that have been distorted, or have had noise added, in ways that have been difficult for computer vision software to solve but easy for humans. According to an Economist article, the team's method was able to "crack nearly half of all CAPTCHAs and one-third of ReCAPTCHAs." The research paper was held back for several months to allow developers time to improve CAPTCHA-generating methods.
Related articles:
Multispectral imaging for robot strawberry pickers
Researchers at the National Physical Laboratory (NPL), UK, have developed an imaging technology to automate fruit-picking robots. The system uses captured images from different wavelengths to determine the ripeness of strawberries before they are picked. The multi-spectral system can safely penetrate a crop's outer layer to identify the level of ripeness against a predetermined standard. The technology can also be used for other applications, such as waste management where it is necessary to differentiate among plastics, paper, and wood. More information is available in a press release.
Computer vision and 3D printing for drug discovery
A fastcodesign.com article reports on an augmented reality (AR) tool for molecular drug design. Arthur Olson of the Scripps Institute Molecular Graphics Laboratory has developed a aystem in which a 3D printer creates solid models of drug and enzyme molecules; and a web cam tracks them to create AR overlays that help scientists figure out how the molecules can fit together.