Performance Metrics and Validation Methods for Vehicle Position Estimators
Adrian Fazekas, Markus Oeser · IEEE Transactions on Intelligent Transportation Systems · 2019
Future applications of the intelligent traffic systems (ITS) will strongly rely on highly detailed traffic data. Beyond aggregated traffic parameters, such as flux, mean speed, and density, used in macroscopic traffic analysis, a continuous location estimation of individual vehicles on a microscopic scale will be required. Different data acquisition techniques based on GPS, radar, and computer vision systems have been already focused in recent years. Although the published studies in many cases include a validation of the developed methods, a more specific analysis on the performance of the vehicle position estimation is necessary. This paper presents a systematic approach to measure the performance of the position estimators based on a comparison of the automatically recorded results to manually generated ground truth of individual vehicle positions in video segments. Furthermore, the methods described in this paper enable a straight forward error handling, which is very useful in the development of position estimators. The concept is exemplified on computer vision-based vehicle tracking.