Adaptive camera to display mappings using computer vision

The video surveillance industry is experiencing dramatic change with the move from analog to digital video. Command centers need to have coordinated viewing of multiple camera feeds at one time, and the ability to switch automatically between feeds and display relevant patterns. Conventional security control rooms include a bank of monitors connected through a switch to an array of security cameras. Fixed protocols are used to cycle the cameras through the monitors, with provisions for human over-ride. Advances in display technology and high speed networks motivate us to propose a radically new model of the human/display interface in the control room. We propose general techniques, based on computer vision algorithms for measuring the saliency of surveillance videos, for mapping video cameras to display space (resulting in variable amounts of display space per camera), and for visualizing the information in each video stream. The computer vision techniques involve statistical characterization of patterns of movements to develop measures of movement saliency (to control the camera to display space mapping), and perceptual modeling of video content to drive the visualization of an individual video stream. We describe a pilot user study to evaluate these ideas.