Don't look now, but Big Brother is just around the corner, say researchers at Texas A&M University. No, not the reality TV program, but automated surveillance cameras that can track people in train stations, the mall and parking lots. Nasser Kehtarnavaz, a professor of electrical engineering and director of the university's digital signal processing program, and his former student Victor Cheng have demonstrated a smart camera that can identify and observe pedestrians in real time. The surveillance system, which they described in the July issue of the Journal of Electronic Imaging, is built around the TMS320C6201 fixed-point digital signal processor on the C62 evaluation module from Texas Instruments Inc. of Dallas. A video board, also made by Texas Instruments, acquires the analog signal from a black-and-white video camera. The signal is digitized and stored on the module and the researchers' detection and tracking algorithm analyzes the feed at approximately 5 to 6 fps. An automated surveillance system tracks four pedestrians on the feed from a black-and-white video camera. Courtesy of Nasser Kehtarnavaz. "The pixel intensities of a current image are compared to those of the background image," Kehtarnavaz explained. "Those pixels that exhibit a difference above a certain level are assigned to moving objects." To ensure that the system concerns itself only with picking out human targets, the algorithm also identifies the features that look like human heads on the vertical axis of the moving objects. Kehtarnavaz and Cheng evaluated the performance of the system by allowing it to observe people in different settings and at different times of the day. Although it nearly perfectly identified and tracked a single person moving through a region of interest or a number of people walking separately, it had difficulties when faced with people walking in a group. I'll be watching you Kehtarnavaz could not comment on any plans to commercialize the system. He did say that it could be used to monitor areas for "suspicious activity." The system needs higher frame rates, however, before it can offer reliable performance in such applications. Eliminating the host PC is the key, Kehtarnavaz said. "For commercial use, dedicated hardware circuitry would be needed to send the images directly to a monitor or to place them on a network," he said. He suggested that dedicated hardware would free up processing time and optimize input/output routines. "At the very least, this would double the frame rate." In the demonstration, the researchers used a "simple video camera," but Kehtarnavaz noted that the algorithm is suitable for integration with other sensors. Infrared cameras in particular would enable the system to track in low- or no-light environments. He suggested, however, that such sensors still might be too expensive for a commercial system. Color, he added, would improve performance but add to the computational complexity.