We present a discrete reversible jump Markov Chain Monte Carlo (rjMCMC) algorithm to build lane accurate maps by solving a partition problem. Our algorithm finds lane middle axes by 2D cubic Hermite splines least squares alignment to trajectory clusters. The steps of the rjMCMC optimization are designed to change the number of clusters and the trajectory assignment. Furthermore, we introduce our tracking approach, which is used to generate our trajectory data base by tracking vehicles in Velodyne LiDAR scans at several junctions in Hannover, Germany. We validate our results by comparing them to a manually generated ground truth map, based on LiDAR mobile mapping data, and a manually generated reference map on base of the trajectory data to evaluate our result by ignoring accuracy errors of the tracking process. Finally, we show the transferability of our approach by applying it to datasets from the map construction data base.
Discrete Reversible Jump Markov Chain Monte Carlo Trajectory Clustering
01.10.2019
1596350 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Fitting multiple projective models using clustering-based Markov chain Monte Carlo inference
British Library Online Contents | 2015
|Sensitivity Analysis of Markov Chain Monte Carlo
AIAA | 2010
|Markov Chain Monte Carlo Modular Ensemble Tracking
British Library Online Contents | 2013
|Markov Chain Monte Carlo for Automated Face Image Analysis
British Library Online Contents | 2017
|