Semi-Automated Video Analysis in Mobile Eye Tracking for the Analysis of Urban Advertising (Abstract)
At the core of our research is the application of computer vision for mobile eye tracking systems which has recently been considered for visual memory augmentation (Roy, Proc. ISWC 2004) and the analysis of scanpath data (Cerf et al., Springer LNAI 5395, 2009). Our study is on the interpretation of eye movement data of about 100 subjects in Austrian public transportation services. We integrated a component for vision based object detection for the semi-automated analysis of huge video and eye movement data for the analysis of pedestrian awareness in urban advertising. We designed a learnable classifier for highly accurate localisation of public electronic displays and propose a strategy for combined manual rejection of false positives. From display detection and the associated broadcasting plan we were able to derive a current content of user awareness and from this a statistical evaluation of eye movements with respect to content categories. The proposed policy for semi-automated object detection automatically covers 63 % of video data with detection accuracy of ca. 98 %. The remaining data are manually evaluated with a minimization with respect to the use of personnel resources so that only 15 % of video data needed full manual interaction.