Image Processing Methods for Computer Vision-Based Fuel Gauge Developed
April 4, 2000
University Park, Pa. --- Taking a "look" at how much fuel is left in the tank could become literally possible now that Penn State computer engineers have developed image processing methods necessary for a computer vision-based fuel gauge.
The developers say a computer vision-based gauge would be much safer than current gauges that rely on a sensor with low voltage electrical leads that come in contact with the fuel. Electrical contact fuel measuring systems are often suspected of having contributed to explosions in aircraft disasters.
The image processing methods were developed by Srivatsan Chakravarthy, who earned his masters degree in August at Penn State, Dr. Rangachar Kasturi, professor of computer science and engineering, and Dr. Rajeev Sharma, assistant professor of computer science and engineering. The work was part of Chakravarthys masters thesis.
In the method the Penn State team developed, nothing need be in the fuel tank except the fuel, Kasturi says. Two transparent glass portholes in the top of the fuel tank allow all the access needed to determine the depth of the liquid inside. One port enables a light source to flash the image of two crossed lines on the surface of the liquid while the other port allows an off-the-shelf video camera attached to a computer to record the position of the lines.
The computers image processing software, developed by the researchers, is trained in a calibration process in which the tank is filled gradually and the position of the crossed lines, as recorded by the camera, are noted at the various depths. The correspondence between the depths in the actual 3-D scene and the 2-D image recorded by the camera is established in a process called digital mapping. The computer can then calculate the change in depth of the liquid in the tank when any one of the units, or pixels, in the cameras image of the crossed lines changes.
The researchers note that the mathematical problems involved in determining depth in a three-dimensional space from a two-dimensional camera image forms a part of almost all computer vision applications. A similar problem has to be solved whether one is equipping a robot with a 2-D camera "eye" to navigate in the real 3-D world or determining the depth of a liquid from its 2-D image.
The researchers used two standard mathematical approaches to solving the problem for the fuel tank and found that one, triangulation, yielded simple, direct solutions that came at low computational cost. Although their computations were performed with a stationary system, they think their approach can be adapted to systems undergoing vibration, turbulence or other displacement.
The researchers conclude that computer vision offers an attractive alternative to the currently available options. They note that their experimental results are available now for studying the feasibility of implementing an actual system that could be incorporated in aircraft, automobiles or other applications.
The project results are detailed in the Department of Computer Science and Engineering Technical Report, CSE-99-006. Copies are available by contacting the researchers.
- Barbara Hale (814) 865-9481 (o)/ (814) 238-0997 (h)
- Vicki Fong (814) 865-9481 (o)/ (814) 238-1221 (h)
- EDITORS: Dr. Kasturi is at (814) 863-4254 or at by email; Dr. Sharma is at (814) 863 0147 or at by email.