Brno University of Technology at TRECVid 2012: Interactive Surveillance Event Detection Pilot

The experiments we describe here involve the interactive surveillance event detection pilot (SED) of the 2012 TRECVid evaluation. Our approach is called Surveillance Network Augmented by Retrieval Event Detection (SUNAR-ED), an open source single-camera computer vision and information-retrieval based surveillance event detection system, which is based on wide area (video) surveillance system (SUNAR). It contains state of the art and experimental methods of object tracking (OpenCV), feature extraction (MPEG-7), search (similarity, spatio-temporal) and learning (Bayes, GMM, HMM, SVM).

We have deployed active learning functionality based on moving objects' trajectory statistics and shape classification using  VTApi (Video Terror Application  programming interface and methodology), a data base interface used for processing and  efficient management,  indexing, retrieval and analysis of image and video data and related metadata, which was created to unify and accelerate the intelligent vision applications development.