A recent academic study delves into the precision of drone surveying without the use of ground control and RTK GNSS correction, with findings now available in a white paper.
The EuroSDR Commission 1 Benchmark conducted an impartial assessment of the genuine geometric accuracy of survey data produced by Remotely Piloted Aircraft Systems (RPAS) across varying control configurations in real-world scenarios.
In the realm of photogrammetry, the primary insights were as follows:
- Depending on the equipment used, accuracies ranged from meter to decimeter levels in both plan and elevation without the support of ground reference points. Notably, the absence of ground data led to compromised accuracy compared to precision.
- Employing RTK GNSS corrections yielded precision and accuracy at the centimeter level, albeit with occasional deviations or outliers.
- Incorporating ground control points (GCPs) marginally enhanced planimetric accuracy over RTK methods. However, the most significant enhancements were observed in vertical accuracy (z-axis).
- The white paper articulates: ‘The integration of GCPs continues to play a pivotal role in pinpointing discrepancies, bolstering the dependability of such surveys.’
These conclusions underscore the imperative of strategically incorporating GCPs in drone photogrammetry surveys. Such points offer calibrated reference coordinates, elevating the caliber of digital mapping and survey outputs.
The survey unfolded in three distinct phases:
- Phase 1: Data processing without the aid of GCPs or GNSS base station information.
- Phase 2: Data processing incorporating GNSS base station details and/or real-time RTK corrections.
- Phase 3: Data processing utilizing GCPs, local GNSS base station information, and/or real-time RTK corrections. Newcastle University was responsible for processing datasets throughout all these phases.
As each phase concluded, the raw data became incrementally accessible to registered participants. Upon completion of each phase, participants were tasked with submitting 3D coordinates for the checkpoints (CPs) and providing a concise overview of their selected processing methodology or software solution.
The benchmark encompassed the evaluation of 41 distinct submissions. These submissions, derived from authentic drone survey data, underwent scrutiny against GNSS-surveyed checkpoints, terrestrial laser scanning (TLS) surveys, and total station elevation readings—all referenced to a national coordinate framework.
The comprehensive findings, alongside the associated white paper, were unveiled towards the latter part of 2023.
The research transpired at a meticulously coordinated test site encompassing autonomous checkpoints (CPs), designated test terrains, and profiles situated within the decommissioned Wards Hill Quarry near Morpeth, Northumberland, UK.
Eight GCPs and 51 CPs were used across the site of 350m x 250m in size.
Data collection for the photogrammetric datasets involved deploying the DJI P1, P4 RTK, and L1 drones at altitudes ranging between 50m and 55m.
Specifications for Each Equipment:
- P1: Features a 45MP full-frame sensor (35.9 x 24mm) with a 4.4μm nominal pixel size. During the survey, this drone operated with automated exposure settings and continuous focus, ensuring images captured had an 80% forward and 70% lateral overlap.
- P4 RTK: Equipped with a 1-inch, 20-megapixel CMOS sensor having a 2.41 x 2.41μm nominal pixel size. The survey configuration maintained a fixed exposure duration and infinite focus, ensuring images had an 80% forward and 70% lateral overlap.
- L1: Comes with a 1-inch, 20MP CMOS RGB sensor with the same 2.41 x 2.41μm nominal pixel size. During the survey, it was programmed to maintain a 50% lateral overlap during its flights.
It’s essential to highlight that post-survey, the DJI Mavic 3 Enterprise has succeeded the P4 RTK. Additionally, the L2 serves as an enhanced iteration of the L1. Both the M3E and L2 offer augmented functionalities, elevating surveying efficiency, accuracy, and precision over their forerunners.
A pivotal finding from the research underscores the advantages of integrating on-the-go RTK and/or localized base station methods, especially when GCPs are incorporated.
As an illustration, the subsequent table, which presents photogrammetric outcomes from the L1, elucidates the enhanced data quality achieved with RTK GNSS corrections. The addition of GCPs further amplified these benefits, particularly enhancing the vertical data metrics.
|Standard Deviation (m)
Findings from the White Paper
The white paper highlights that when the L1 was exclusively processed as a photogrammetric dataset, it achieved centimeter-level accuracy, approximately 1x GSD in the horizontal plane and 1.5x GSD in vertical measurements, especially when GCPs were integrated. Notably, the L1 managed to achieve a Ground Sampling Distance (GSD) of 0.018 m from an altitude of 50m.
Furthermore, the research emphasized the superior performance of the P1 in the realm of drone photogrammetry, especially when juxtaposed with its predecessor, the Phantom 4 RTK.
To illustrate, the DJI P1 exhibited fewer discrepancies when aligned against TLS data, presenting a more consistent output compared to the P4 RTK, as depicted in the accompanying graphic.
Key Findings from the Study:
- P1 vs. P4 RTK Without Ground Control or RTK: The P1 achieved accuracies at the decimeter level in both plan and elevation without the support of ground control or RTK, in contrast to the P4 RTK, which exhibited accuracies at the meter level, albeit with a few anomalies.
- Planimetric Accuracy in Different Phases: During Phases 2 and 3, with the aid of RTK corrections and GCPs respectively, the P4 RTK showcased a planimetric accuracy around 0.020m (1x GSD), while the P1 outperformed with a finer GSD of 0.007m when GCPs were employed.
- Point Cloud Density Observations: The P1 registered a point cloud density of 1310 pts/m² on flat asphalt terrain and 1320 pts/m² on exposed rock devoid of vegetation. In comparison, the P4 RTK captured a point cloud density of 200 pts/m² across identical surfaces.
White Paper Conclusions on Photogrammetric Surveys:
The white paper elucidates various insights derived from multiple participant submissions:
- Phase 1: In the absence of ground control or localized base station data, submissions primarily yielded solutions at the centimeter to decimeter range, showcasing a notable disparity between accuracy and precision.
- Phase 2: With the integration of RTK GNSS corrections, a more balanced accuracy and precision emerged at the centimeter level. However, inherent datum inconsistencies in both plan and elevation became apparent, especially without GCPs, as observed in Phase 3.
- Phase 3: The utilization of GCPs culminated in optimal precision and accuracy, approximately 1x GSD, marginally surpassing the results from Phase 2.
- General Observations: Throughout all phases, anomalies persisted, underscoring the criticality of meticulous processing procedures encompassing camera calibration, coordinate transformations, and the like. Especially in today’s context of advanced “black box” SfM photogrammetry techniques, GCPs remain instrumental in pinpointing and rectifying such discrepancies, thereby bolstering survey reliability.
LiDAR Findings from the Study
The research delved into comparing the quality of LiDAR data derived from the DJI L1 and the Riegl MiniVUX, the latter being integrated with the DJI M600 drone for this evaluation.
DJI L1 Specifications:
The DJI L1 is equipped with a Livox LiDAR scanner, complemented by a GNSS and inertial navigation system. Each flight entailed 13 parallel flight paths, maintaining a 50% lateral overlap. Operations were conducted at an altitude of 50m above the terrain, utilizing a 160 kHz sampling rate, and encompassing a repetitive scanning sequence that allowed for a maximum of three echo returns.
Riegl MiniVUX Specifications:
The Riegl MiniVUX setup comprises dual GNSS NovAtel receivers/antennas, an IMU SPAN Airborne EKINOX by SBG Systems, and a dedicated Riegl MiniVUX-1 LiDAR module. This system was flown at an altitude of 60m from the launch site, moving at a speed of 6 m/s. The scanning pattern included five parallel routes and two perpendicular cross strips, capturing data with a swath width of 280m at a 20% reflectance level.
The data analysis revealed that the DJI L1 exhibited a more robust point cloud density compared to the Riegl MiniVUX. This distinction is further elucidated in the subsequent table, showcasing representative samples that delineate the disparity in point cloud densities between the two systems.
|1 ( flat terrain on asphalt)
|5 (bare rock, no vegetation)
DEM Resolution Comparison:
The DJI L1 demonstrated its capability by generating a Digital Elevation Model (DEM) with a spatial resolution of 0.05m. In contrast, the Riegl MiniVUX yielded a DEM with a slightly coarser spatial resolution of 0.125m.
Interestingly, despite the Riegl MiniVUX offering a less refined spatial resolution, it exhibited fewer discrepancies or errors compared to the DJI L1.
To provide a tangible example, the accompanying visual representation indicates that in a cross-sectional view of Area 5, the DJI L1’s point cloud was situated approximately 0.15m below both the TLS data and the point clouds generated by the Riegl MiniVUX.
White Paper Insights:
The white paper highlights: “A significant factor to consider is that the R1 GNSS base station was utilized for processing the trajectory data from the Riegl MiniVUX RPAS. This could potentially elucidate why it exhibited superior accuracies in comparison to the commercial RTK capabilities of the DJI L1.”
Additionally, the document points out that during the period of data processing, the DJI Terra software lacked the functionality to incorporate Ground Control Points (GCPs). However, it’s pertinent to note that this capability has since been integrated into the software. As a consequence, the accuracy of the DJI L1’s output was solely contingent upon the RTK corrections facilitated by a commercial service.
Furthermore, it’s essential to factor in that Riegl solutions often command a higher price point relative to DJI payloads, making them a premium investment in the context of surveying tools.