Cumulation of walking paths and detection of anomalies

Abstract

When people walk, there appear paths. When there are paths, people walk there. In this study, we introduce a simple vision-based system that traces a walking people and running vehicles, identifies the paths and determines the anomalies (outliers, abnormal motions) departing from the paths. The system obtains video data at the location above some meters or ten and some meters above the ground. An optical flow method is used to trace the moving object. Initially, the system knows no road. By cumulating trajectories of moving objects, the system assumes the set of the trajectories as a path. Once paths are assumed, a motion not following the roads is detected as an abnormal motion. When the motions similar to the abnormal motion continue, the system reconsider that there appears a new path. Appling to the real outdoor environment, we have verified the validity of the present method. © 2012 The Japan Society of Mechanical Engineers.