A Semi-Automatic Data Annotation Tool for Driving Simulator Data Reduction
Chris Schreiner, Harry Zhang, Claudia Guerrero, Kari Torkkola, Keshu Zhang · 2007
Manually annotating large video and digital databases of driving behavior is costly and time-consuming. In this paper we discuss a data annotation tool that automates the process and reduces the number of man-hours required to annotate data. Our laboratory has utilized this tool on a large database of simulated driving data to develop context aware driving systems. Our semi-automatic data annotation tool supports our research efforts for driving database creation to enable data-driven approaches in the driving domain such as driving state and manuever classification. The annotation tool employs Random Forests as bootstrapped classifiers which are then used to predict annotations for new data files. We describe an experiment which generated a large database of driving data with our DriveSafety simulator, the process by which annotations are automatically generated, and the results of how using the data annotation tool markedly reduced the amount of time required to annotate the data among three users with varying levels of annotation experience. Our major contribution in developing this tool is making parts of the annotation process automatic enabling the user to verify automatically generated annotations, rather than annotating from scratch. This tool has the potential to become a standard data reduction technique.