VicarVision develops video analytics solutions for robust real-time detection of moving people for day-night surveillance by analyzing their motion patterns, based on expertise from the project partner TNO (the Netherlands Organisation for Applied Scientific Research). With our project partner the implemented solutions will serve as clever surveillance services on-camera, on-server and in-the-cloud. Examples of these services are people counting, virtual fencing, heat mapping, and loitering detection.

Recognition of Moving Humans

TNOSBIR Using a level set method silhouettes are extracted from the motion in a video stream. These silhouettes are classified as human or non-human by a comparison with silhouettes from the CASIA Gait Database with 1102 examples of persons. A silhouette score is computed, which is a measure for the similarity of the silhouette with the human silhouettes of the CASIA gait database. Also, based on the optical flow patterns of the silhouette tracks, an optical flow score is computed, which is a measure for the similarity of these patterns with optical flow patterns typically for humans. The final classification as human or non-human is based on combining both scores.