Memory-Based Attention Control for Activity Recognition at a Subway Station
Karl Fredric MacDorman, Hiroshi Nobuta, Satoshi Koizumi, Hiroshi Ishiguro · IEEE Multimedia · 2007
We have developed a multicamera system, Digital City Surveillance, which uses a new calibration-free behavior recognition method for monitoring human activity at a subway station. We trained nine support vector machines from operator-classified data to recognize 512 combinations of events. Our method of attention control greatly reduced computation and increased classification accuracy