Video Analytics

In the area of video analytics, our group conducts research and develops novel computer vision algorithms (also based on deep learning and other machine learning tools) for various problems such as scene text detection and recognition in natural videos and images, video segmentation, visual recognition and scene understanding, object tracking, and more. A special focus of our group's research work is on effective analysis of video captured by moving platforms, also in combination with geographical information (GIS), for applications such as dashboard and body-worn camera analytics.

Video text extraction:

Text contained within natural images and video can be of great semantic value, and therefore detection and recognition of such text is an important task in computer vision. Our group conducts research aimed at automatic detection and recognition of text in videos and images, including text in natural scenes, using state-of-the-art technologies. The recognition system also provides localization information in the form of regions in the image that contain text, and is used by Watson to perform visual recognition of text.

Video Scene Detection:

Since many videos consist of a number of contextual scenes, it is important that algorithms for cognitive understanding or analytical deduction be applied to one scene at a time. In this respect, our group researches the area of temporal video segmentation, where automatic parsing of a video into comprehensive scenes is obtained. High quality decomposition of a video into its scenes for a variety of genres can be a difficult but extremely important task.