Rolling Shutter Effect Important in UAV Photogrammetry

Many are not aware that photographs taken with a UAV are subject to the rolling shutter effect. (Be sure to watch the video at the end for great visual explanation.)
The main disadvantage of UAV photogrammetry is that it uses non-metric cameras. These cameras have low-quality movable lenses, a small focal length, and most have a CMOS sensor with an electronic rolling shutter.  Each scanline of the acquired image with a rolling shutter is exposed at a different time, which induces image blur when the camera is in motion.  The rolling shutter effect becomes significant when flying the drone fast and/or when flying at low altitudes.

Some UAV photogrammetric software has the option to mathematically model this distortion. This study will analyze the rolling shutter effect and the effectiveness of these software in removing this systematic error against a control field with many check points.

Project Area

The project area was located west of the state of Illinois with approximate dimensions of 460 m x 1000 m. The terrain was flat, containing light vegetation, scattered buildings, and sports fields. The site had 97 GCPs (ground control points) mostly evenly distributed throughout (Figure 1).

Flight Parameters

The photography was acquired using “Ricoh GR Digital 3” camera mounted on a Trimble Gatewing X100 (Figure 2) fixed-wing UAS with a single frequency (L1) GPS receiver.  An average GSD of 5 cm was achieved from 530 feet above ground level (AGL) and images were captured with 80% forward overlap and 60% side overlap.

Software

The processing of the data was performed using three different commercial software. Software I does not have the option to model the rolling shutter.  Little information is provided on how software II and software III model the rolling shutter effect.

Data Processing

This data set was advantageous because it contained a large number of GCPs (97). Five control point layouts were chosen as described in Table 1. They are:

• Control layout A (15 GCP)
• Control layout B (13 GCP)
• Control layout C (11 GCP)
• Control layout D ( 9 GCP)
• Control layout E ( 5 GCP)

The data were first processed in all 3 software with rolling shutter off for each control layout.  Second, the data were processed in software II and III utilizing the option to correct for rolling shutters.

Data Analysis – Rolling Shutter Effect

a. Shutter OFF Results

The planimetric and elevation errors for layout A are shown in Table 2 as an example.  Software I and III exhibited an outward planimetric systematic error from the center of the block for all control schemes.  Software II showed this effect to lesser extent.  All software showed some doming effects in the z-direction. While all software met the expected elevation accuracy (10 cm) beyond 5 GCP’s, I and III failed to achieve the planimetric accuracy claims (5 cm) as shown in Table 3.  Software II achieved planimetric accuracy for 5 GCP’s while elevation accuracy degraded.

b. Shutter ON Results

The planimetric and elevation errors for control layout A are shown in Table 4.  A summary of the results is shown in Table 5.  Elevation accuracy was not consistently achieved within Software III overall and planimetric accuracy degraded when 5 control points were used. Both software II and III exhibited some systematic errors for the 5 control point layout.

Conclusion

The rolling shutter effect in UAV photogrammetry needs to be modeled and cannot be eliminated by increasing the control points.  Different software produce different results in attempt to eliminate this systematic error.

###

Riadh Munjy, Ph. D., P.E.

Professor of Civil and Geomatics Engineering

riadhm@csufresno.edu

Jacob Lopez

Graduate Student

jakelopez1@mail.fresnostate.edu

California State University, Fresno

Note – If you liked this post click here to stay informed of all of the 3D laser scanning, UAS, autonomous vehicle and Lidar News. If you have an informative 3D video that you would like us to promote, please forward to editor@lidarnews.com.

Figure 1

Figure 2

For the remainder of the tables click here as the size of the tables were too large to be imported.