Improving UAV-SfM time-series accuracy by co-alignment and contributions of ground control or RTK positioning

Abstract

Unmanned Aerial Vehicle Structure from Motion (UAV-SfM) photogrammetry is increasingly applied to topographic change detection, which requires multitemporal Digital Surface Models (DSMs) with high relative accuracy. Of these tools, Ground Control Points (GCPs) and an image processing method called co-alignment have so far shown promising results for change detection studies. However, there is still insufficient research on the extent of improving 3D model accuracy by combining these tools. In our study we assess absolute and relative accuracy of 120 DSMs generated through 24 workflows of UAV-SfM photogrammetry. Surveys were acquired with two different UAVs with Real Time Kinematic (RTK) or generic Global Navigation Satellite System (GNSS) positioning, and processed with varying combinations of survey co-alignment and GCPs. We show that co-alignment reduces relative errors to below 2 cm regardless of positioning quality. A single RTK survey in a co-aligned project is sufficient to obtain high absolute xy accuracy, but GCPs for at least one survey are still required to reduce absolute z error. We demonstrate that co-aligning RTK surveys with generic GNSS surveys results in RTK class accuracy for all surveys, even when mixed sensor grades are used. Our findings enable high-accuracy change detection with lower accuracy archived images when combined with RTK surveys. For future UAV-SfM change detection studies, we recommend to apply co-alignment for all studies, and where possible to include GCPs and RTK image coordinates in one survey to optimize absolute accuracy. Collecting and digitizing GCPs in multiple surveys has shown little additional benefit when co-alignment is applied and therefore may be omitted to save time, especially in challenging field conditions.