Mapping with PPK, An Assessment of Accuracy

Summary:

The PPK GPS system as integrated into the E384/6 is capable of delivering high scale accuracy when used alone and high absolute accuracy when a known good point is available near the flying site. In this case study, we produced an orthomosaic in the Drone Data Management System™ with 2.45cm horizontal and 5.08cm vertical RMSE, as compared with a survey-grade GPS on the ground.

Data Collection, Equipment and Processing:

For this study, a high resolution orthomosaic was created and multiple ground control points (GCPs) were collected to check the results. The orthomosaic was built using only the geotags from the onboard GPS.

The orthomosaic was built using an E384 outfitted with an Emlid Reach GPS Receiver and a Sony QX-1 Camera. The QX-1 was modified by Event 38 to provide the time that each image was taken down to one millisecond. Having this accurate timestamp is what allows the geotags, and subsequently the orthomosaic to be reconstructed with a high degree of accuracy. After the flight, we processed the RINEX file using an Event 38 utility and RTKLib postprocessing software. These produced a geotag file with coordinates for each image. Then the images and geotag file were uploaded to the Drone Data Management System™ to build the orthomosaic and DEM.

The GCPs were measured using a Trimble R6 Model 4, with corrections provided by the Ohio Department of Transportation (ODOT) VRS network. On the day of the flight, the Trimble R6-4 was not present. Instead, the Emlid Reach base station was positioned on a marker that had been previously recorded.

Once all the data was compiled, the orthomosaic was visually inspected and each feature representing a GCP was marked manually. Finally, we measured the distance between each marked GCP and the actual GCP location as measured on the ground.

Results:

Table 1 contains the full list of errors for each GCP, both horizontal and vertical. What we found is some variance in individual measurements, with up to 4.29cm horizontal and 6.7cm vertical error. Overall the RMSE was 2.45cm horizontal and 5.08cm vertical.

Table 1 - Errors between GCPs as measured from the orthomosaic vs on the ground
GCPHorizontal Error (cm)Vertical Error (cm)
11.376.7
22.496.5
34.295.6
41.733.9
52.185.4
62.164.8
73.842.1
81.633.0
91.445.1
101.566.1
111.122.8
123.156.3
RMSE
2.45cm
5.08cm

Conclusion:

The errors determined from this case study should be considered as characteristic of the Emlid Reach system when used with an E384/6 and QX-1 for measurements relative to the base station. In this case, we positioned the base station at a known good coordinate, so we were able to match the absolute GPS coordinates collected by the Trimble R6-4 very closely. When using the Emlid Reach by itself, even when using an outside correction system such as the ODOT VRS network, it is still possible for a significant offset to be present. The data will be accurate relative to the base station, but an offset between the Reach’s calculated base station coordinate and the true coordinate may exist. If the base station position is marked, it is possible to correct the absolute position at a later time.

At least some of the variance in error is due to the difficulty in locating certain markers accurately at the collected 2.5 cm/pixel ground sample distance (GSD). Although it is likely possible to improve this by collecting data at a lower altitude, we felt it was important to characterize the accuracy at a normal flying altitude which is more typical for our customers working in surveying, mining, and agriculture.

Based on this data, with RMSE errors of just 2.45cm horizontal and 5.08cm vertical, we feel confident that the PPK GPS system as implemented in the E384 and E386 can deliver reliable scale accuracy when used by itself and absolute accuracy when used in conjunction with a known good coordinate.