Comparing the senseFly S.O.D.A. and S110 Sensors – Part 2.
Audience: Pilots, Mission Planners, Post Processing
The senseFly S110 RGB sensor has been a reliable part of our UAS processes. Coupled with the eBee RTK, the S110 sensor has proven capable when used with proper lighting and meteorological conditions.
In late 2016, senseFly introduced their new S.O.D.A. sensor for the eBee platform. This sensor has some material improvements over the S110 RGB and this brief is a summary of the initial findings on the operational use of the S.O.D.A. sensor and integrating it into the business processes.
This high-level overview is provided as guidance mainly for our pilots, mission planners and photogrammetrist, but should be considered by all team members. In the previous Part 1 brief the S.O.D.A. sensor was examined at the hardware level, how the changes impact mission planning, and comparison of the actual raw imagery from the flights. Readers will find it beneficial to review Part 1 prior to continuing. Read it here.
As Part 1 was focused on the implementation and operational aspects of the two sensors, Part 2 (this brief) of this series will examine the resulting orthomosaic quality produced from the imagery of the two sensors. Since Pix4D is recommended by senseFly, we will use Pix4D version 3.1.23 as the post processing software. The resulting point cloud will be compared in the next brief (Part 3).
Each flight conducted in Part 1 was post flight processed in eMotion 3 and a Pix4D package was created by eMotion for each flight’s data. This post flight process by eMotion creates a basic Pix4D project file, verifies the images, modifies the image metadata to include the corrected GPS information and camera positional information.
Each eMotion 3 created project is loaded into Pix4d using the 3D Maps template with the following deviations made to the processing parameters of all three test projects;
Calibration set to Accurate Geolocation,
Image Scale set to Original size,
No generate textured mesh,
Matching Window set to 9x9,
So basically, a point cloud and an Ortho are the only outputs to be generated for the projects in this brief.
Table 1 below shows the absolute camera position and orientation uncertainty for all three flights. The offset for both Mean and Sigma values are smaller for the S.O.D.A. payload than for either S110 flight. Interestingly, the uncertainties get smaller for the second S110 flight which was conducted at about the same altitude as the S.O.D.A. flight (~400ft ATO).
Table 1 – Absolute Camera Position and Orientation Uncertainties by Pix4D.
In Image 1 below the computed camera positions are shown with uncertainty ellipses (magnified 100x) on the right / bottom for each data set (from Pix4D).
Image 1 – Pix4D Computed Image Positions with Uncertainty Ellipses
While all potential error values above are relatively small, the S.O.D.A. sensor’s computed uncertainty data shown has much lower uncertainty than either of the S110 data sets.
Pictured below (Image 2) is the orthomosaic derived from the mission. This image is provided to orient our readers to the mission area and provide visual reference for our discussion. This brief will analyze the quality of the orthomosaic as rendered, therefor no analysis of absolute accuracy will be analyzed herein. The Z axis information will be analyzed in the point cloud discussion in the next brief (Part 3). In the image below there are areas marked in red that will be the focus of comparison between the two sensors. The areas of interest are labeled A through I. Each area marked will be extracted at a high zoom for easier comparison viewing.
Image 2 – Ortho of Mission with Marked Comparison Areas.
In Area A (image 3) the green building that was used as an exhibit in Part 1 image analysis is shown. The S.O.D.A. ortho added a ghost building, added shadows to the roof and eliminated a small shed area (See arrows). The S110 shows the proper placement of all structures, and did a better job of color correcting the roof than our sample image from Part 1 would indicate.
Note: The author was present on location and did not witness an apparition of a building. Perhaps it was there, as in the Star Trek episode “The Tholian Web”, drifting between universes and the SODA captured it?
Image 3 – Ortho of Area A.
In Area B (image 4) the ortho images are comparable in quality. However, the S110 ortho has slightly better color and contrast.
Image 4 – Ortho of Area B.
In Area C (image 5) the ortho images are not comparable. The S.O.D.A. ortho placed trees along the road bed (visible in grey) where there are none. The S110 ortho also expanded the canopy over the road (visible in grey) but not nearly as inaccurately as the S.O.D.A. image. Ground inspection shows the only tree on the left side of the road is a deciduous tree (circled in Red) and lone conifer just below this area. The S.O.D.A. ortho in this area is just plain inaccurate.
Image 5 – Ortho of Area C.
In Area D (image 6) the ortho images are also not comparable. The S.O.D.A. ortho placed shadows of trees, apparently projecting copies of existing shadows into the green pasture area. The S110 ortho does not contain the same shadows in this area. The S110 ortho also seems to better render the detail of the dormant grass areas blended into the greening areas.
Image 6 – Ortho of Area D.
In Area D (Image 7) the ortho images are not comparable. The S.O.D.A. ortho placed shadows of trees, apparently projecting copies of existing shadows into the grass area. The S110 does not contain the ghosted shadows.
Image 7 – Ortho of Area E.
Area F & H
In Area F (Image 8) and Area H (Image 9) the S.O.D.A. ortho contains a clearly stitched tree top where there is no tree. The S110 ortho more accurately shows the location of the tree top and the green area is grass surface. The S110 ortho clearly does a better job of blending the edges when stitching areas of dissimilar color.
Image 8 – Ortho of Area F.
Image 9 – Ortho of Area H.
In Area G (Image 10) of a solitary tree, the S110 ortho images render a much clearer definition of the tree and canopy limbs. This tree is a large tree of approximately 85ft in height. The S.O.D.A. ortho of the same tree is unfocused and does not appear to have the same structure or size as does the S110 version.
Image 10 – Ortho of Area G.
In Area I (Image 11) of a boundary tree line, the S110 ortho renders a much better merge of the canopy edge along the field edge. The S.O.D.A. ortho of the same canopy edge contains harsh / sharp stitch lines and also covers a pit with an image of a tree top (right arrow). This pit is set well off from the tree line and is more clearly visible in the S110. However, in this area of the ortho, the S.O.D.A. version appears to have better focus and color definition.
Image 11 – Ortho of Area I.
Reviewing the orthomosaic segments attached in this brief, the orthomosaic derived from the S.O.D.A. images contains areas that are unreliable, and are in some areas, are just plain incorrect representations of the surface features below. The orthomosaic generated by the S.O.D.A. sensor in this test tract would not usable for production work. The orthomosaic produced by the S110 in both flights (1.2in and 1.8in GSD) are both accurate and, when considering the quality of some of the raw images analyzed in Part 1, are surprisingly well color balanced.
These issues with the S.O.D.A. are likely not the fault of the camera, as it's image output quality were better overall as discussed in Part 1. It is likely a fault within Pix4d software, being a new sensor with new lens and sensor geometry. It may take a period of time until Pix4D can catch up with this product.
However, until that time, not use the S.O.D.A. sensor payload for orthomosaic work. Mission planners should be aware of this and specify whether or not an orthomosaic will be required output on a given mission so that the S110 can still be deployed.
In the next brief, Part 3 of this series, the point cloud data from the flights will be examined.
Fly Safe and have Fun!