The project
This was a data processing project in Jammu and Kashmir, not a flying one. A drone survey company held the contract from a government survey organisation, and its four or five pilot teams captured data with around ten drones. The client’s own survey team captured data as well. The processing was subcontracted to the team I manage.
The scope was about 1,500 km² of data processing, split into 237 blocks across 12 towns. I manage the operations and the data processing team, and the work is delivered as an orthomosaic, a DTM, a DSM and a point cloud.
What was delivered
Processed areas were submitted to the client for approval.
Orthomosaic
One georeferenced, distortion-corrected image of the whole area, stitched from the drone photos.
Shows what is on the ground.DTM
The height of the bare ground, with trees and buildings removed.
Shows how the ground rises and falls.DSM
The height of the top surface: trees, roofs and everything standing on the ground.
Shows what stands on the ground.Point clouds
The 3D measurements the other products are built from.
Source data for checks and re-processing.Inside the client’s premises
The area is security-sensitive, so the client did not want the data to leave its premises. It provided a space on site, and we set up four high-end processing systems there. All of the processing was done inside.
A slow first month
June was slow. Setting up the systems took time, and the data raised problems that had to be understood before processing could run properly:
- Data management. Storage was a problem at this volume.
- Many sources. The data came from several teams at several sites, and it had to be separated and checked before it could be processed.
- Ground control. The ground control points were blurred in the imagery, so the centre of each marker could not be found. Without usable control, aligning the photos was very difficult. The data had been captured before my team arrived, and the area is security-sensitive, so a reflight was unlikely.
Finding and sorting out these problems took the whole month. The team processed 24 km² in June.
How the problems were solved
Two changes made the difference.
- Post-pointing with DGPS. The team went back to the site and surveyed new points that could be seen clearly in the photos, and used those as control points. This solved many of the alignment problems.
- A validation funnel. Before data went into processing, it was checked: was any of it missing, were the ground control points fully visible, and was the whole flight plan covered? Setting up this funnel, and the pipeline behind it, took time at the start, because data was arriving from several teams at several sites.
Month by month
Once the validation pipeline was running, output rose every month.
September’s 696 km² was about 29 times June’s 24 km². By October, the project had processed about 1,500 km² in total.
Why I am writing it up
This is the biggest milestone of my career so far. Samara Geospatial processes drone and lidar data into survey-ready outputs, and this is the kind of work we want more of: large areas, data from many teams, and a deadline. If you have captured data waiting to be processed, tell us about it.