Outdoor Ground Truth Estimation via GNSS Carrier Phase and Factor Graph Optimization

Kurzfassung

Assessing the quality of a positioning solution is essential for validating high precision GNSS algorithms. Generating a reliable reference trajectory is often challenging because it usually requires expensive sensor units, careful calibration, and independent validation. This paper evaluates several estimators for precise GNSS based trajectory generation using an undifferenced multi frequency PPP float observation model. The compared estimators are an error state Kalman filter (ESKF), a Rauch Tung Striebel (RTS) smoother, sliding window factor graph optimization (FGO), and full batch FGO. The objective is to identify which estimator provides the highest post processed precision from the available GNSS data while also highlighting the practical drawbacks of estimators, especially in terms of computational cost and accuracy. All implementations use the same data handling, observation model, stochastic settings, process model, and float ambiguity handling. The benchmark uses an open sky kinematic GNSS data set of 6000 epochs sampled at 2 Hz. A two epoch sliding graph behaves close to the forward ESKF, a 30 epoch graph provides measurable improvement, and the full batch graph gives the strongest post processed trajectory. The study clarifies the accuracy and cost tradeoffs of FGO and shows why full batch optimization is a suitable reference quality post processing estimator for trajectory generation with GNSS.