Person-on-person violence detection in video data

Abstract

We address the problem of detecting human violence in video, such as fist fighting, kicking, hitting with objects, etc. To detect violence we rely on motion trajectory information and on orientation information of a person's limbs. We define an Acceleration Measure Vector (AMV) composed of direction and magnitude of motion and we define jerk to be the temporal derivative of AMV We present results from several data sequences involving multiple types of violent activities.