ESRI Summit 2013
I have been fortunate enough to be part of the ESRI South-Africa Summit 2013 held at the Champagne Sports Resort. I had the opportunity to fly the ESRI UAV there and see it as the highlight of all my efforts in this blog. I could not only do what I love, but also had the opportunity to finally test the results against proven equipment.
I could do a few flights photographing both the front lawn and a part of the golf course. I was then assisted by Ian from Leica to capture a set of control points on the lawn using their excellent GPS hardware. The accuracy of the Leica GPS was in the region of 4 to 8 mm when making use of differential GPS! We also captured a whole host of cross-check points to test the final results against. Without Ian’s help, I would not have been able to get such a good quality set of points to test against.
I then built an ortho photo and a digital elevation model using a trial of Agisoft’s excellent Photoscan software. I imported the GPS measured control points and cross-check points into ESRI’s ArcMap and measured the differences between my Digital Elevation Model and the measured points. I think the trial has made it’s point and I hope to acquire a licence of this software in the very near future.
I was stunned by the results! Height accuracy of the model was between 25mm and 95mm in the control point area. Accuracy was between 40mm and 400mm in the area outside the control points. The worst accuracy was at the edges of the photographed area – as expected.
These areas would typically not be used as there is a lot of distortion there due to that area being made up mostly of oblique photos. Other large errors was due to the stereo process measuring the height of vegetation and not of the ground under the vegetation. This is a limitation of this type of height model generation that is not present using traditional methods and advanced methods such as LAS mapping.
Even though the process has limitations, it should be considered against the cheap cost of the platform compared to traditional hardware. It should also be noted that the results were obtained using low quality runs on the model to keep the processing time low, enabling us to get results in less than a day. Also of note, is the fact that the plane was flown manually and without the assistance of the auto pilot as I am still awaiting the replacement auto pilot after the servo failure and crash I had last year.
The week end yielded a total of 9GB of video and photos taken from the UAV. All in all a very fun and productive week.
Here is a compilation of some of the video taken from the plane: