Assessing geologic image interpretations errors occurring in extraterrestrial robotic exploration

Abstract

Measuring the error introduced by the machine and human in a man-machine system is an important step in understanding and improving the system. Two experiments assessed the error from the total system in length measurements and shape classification. A separate analysis calculated the error introduce by the camera. The average error for length measurements was 2.33 pixels with a standard deviation of 2.399. The error in edge location introduce by the camera was 5-10 pixels. This indicates that humans generally are good at measuring lengths, but have difficulty near the threshold of resolution. Results from the classification task indicate that humans have lower then expected intra and inter geologist consistency. The study indicates that increasing the resolution of the camera system increases length measurement accuracy and that training may be the best way to improve classification consistency.