In July 1995 in Chattanooga, Tennessee, a clerk in a convenience store was shot and killed during a robbery. The images on the videotape from the store surveillance camera were frustratingly fuzzy. So the police detectives investigating the crime sent the grainy footage to ORNL.
Using an early version of a software tool that forms a sharper image by extracting and compiling information from multiple images of the same subject, a group in ORNL's Instrumentation and Controls (I&C) Division were able to show the police a clearer picture of subtle differences in the crime scene. As a result, the detectives could see the suspect's foot and a muzzle flash from the suspect's weapon, showing that he had fired the gun in the store. Identification of the muzzle flash discredited the suspect's story to the police that the weapon had gone off accidentally in a back room scuffle with the clerk. The new evidence led to a guilty plea, a conviction, and a sentence of life in prison without parole. In this case, the ORNL technology also helped avoid a death penalty trial that would have cost Tennessee taxpayers $100,000 with no guarantee of a conviction.
ORNL's Video Imaging Tool for Aiding Law Enforcement (VITALE) does more than detect subtle changes in surveillance data. It doubles the quality of the videotapes, which are often fuzzy because they are recycled perhaps hundreds of times. Using digitized frames of analog videotape, VITALE algorithms and other techniques sample multiple views of the same subject and "fuse" the video frames together to generate a higher-resolution image. Ken Tobin, leader of the I&C Division's Image Science and Machine Vision Group, who developed the technology with group members Tom Karnowski and Tim Gee, says the technique makes it possible "to get more pixels out of the data so that facial features are sharper and license plates are more legible." The multi-frame fusion technique, whose development is being funded by the Department of Energy, has enabled the researchers to make out words and spots on T-shirts and details of facial features and other identifying marks.
The ORNL software package will be tested by the U.S. Secret Service in mid-1999 and released commercially to local and federal crime fighters by 2000. Besides law enforcement applications, it also could be used to improve the resolution of medical and satellite images.
Originally published by Oak Ridge National Laboratory.
By e-mail SCP@notbored.org
By snail mail: SCP c/o NOT BORED! POB 1115, Stuyvesant Station, New York City 10009-9998