Fully-automatic annotation of scene vidoes: Establish eye tracking effectively in various industrial applications

Modern mobile eye-tracking systems record participants' gaze behavior while they move freely within the environment and have haptic contact with the objects of interest. Whereas these mobile systems can be easily set up and operated, the analysis of the stored gaze data is still quite difficult and cumbersome: the recorded scene video with overlaid gaze cursor has to be manually annotated - a very time consuming and error-prone process - preventing the use of eye-tracking techniques in various application fields. In order to overcome these problems, we developed a new software application (JVideoGazer) that uses translation, scale and rotation invariant object detection and tracking algorithms for a fully-automatic video analysis and annotation process. We evaluated our software by comparing its results to those of a manual annotation using scene videos of a typical day-by-day task. Preliminary results show that our software guarantees reliable automatic video analysis even under challenging recording conditions, while it significantly speeds up the annotation process. To the best of our knowledge, the JVideoGazer is the first software for a fully-automatic analysis of gaze videos. With it, modern eye-tracking techniques can be effectively applied to real world situations in various application fields, like research, control, quality management, human-machine interactions, as well as information processing and other industrial applications.