Virtual Ground Truth in Vehicular Sensing Experiments: How to Mark it Accurately

Road surface quality monitoring is an important requirement for efficient, safe and comfortable transportation. However, the data collection is made difficult by the scope of the data source. Therefore, participatory sensing is a promising approach for road damage assessment. We are developing a vehicular participatory sensing application using Android smart-phones for pothole detection. This paper describes lessons learned from our field tests, which have exposed the deficiencies in terms of collected data quality. Nevertheless, the tests provide invaluable experience for planing future field tests and improvements to the test execution procedure for vehicular sensing researchers. Based on empirical and analytical results, we conclude, that semi-automated ground-truth reference point recording by a human observer in a moving vehicle while doing the actual data collection is imprecise as a consequence of multiple technical and human factors. We also discuss the motivation, why careful pothole position marking and categorization by walking along the test track is capable of providing highly accurate ground-truth. Keywords-real-world experiment experience; data quality; vehicular sensing; participatory sensing; Android OS.