Pix4D participated in a UAV training
mission in Nepal, teaching students how to
use drones and image-processing software
to create professional maps and models
for a wide range of humanitarian and
development purposes
On April 25, 2015, a massive earthquake of 7.8 magnitude struck Nepal. It killed over 10,000 people, shook the lives millions and left many homeless. On May 12, two more devastating quakes, of 7.3 and 6.3 magnitude struck, further killing and injuring people while disrupting the ongoing rehabilitation operations. With millions rendered homeless, roads destroyed and a major part of the tiny Himalayan nation in a mess, once the immediate rescue and rehabilitation was over, Nepal needed urgent restoration work. In this background, Pix4D participated in a week-long UAV training mission in Nepal in September 2015, teaching engineering students at Kathmandu University how to use drones and image-processing software to create professional maps and models for a wide range of humanitarian and development purposes.
Although drones and photogrammetric software for mapping applications are already being used in the fields of surveying and geomatics, solutions are quickly developing for other fields, such as emergency response. In the event of an earthquake like the one that struck Nepal, maps and models produced from drone-acquired imagery and image-processing software can help assist search and rescue operations, damage assessment, reconstruction, preparedness planning and cultural preservation.
The project
Before the training, Kathmandu University (KU) had already been conducting research with drones but lacked the resources and training needed to expand its expertise. Humanitarian UAV Network founder Patrick Meier spearheaded the drone-mapping training in collaboration with KU’s Department of Civil and Geomatics Engineering, Kathmandu Living Labs (KLL), Pix4D and UAV maker DJI, with the intent of building a community of Nepali UAV operators skilled in imagery analysis. By spending time with the leading experts in drone technology, students and young professionals learned best practices, guidelines and regulations regarding drone operation. They also received hands-on flight instruction from DJI and software training from Pix4D. Trainees created specialized flight plans using Pix4Dmapper Capture app for image acquisition, then input that data into the image-processing software Pix4Dmapper to create precise 3D models and maps for further analysis. In an emergency response scenario, these kind of maps and models provide critical information for disaster relief. Although satellite imagery has been used in these situations for decades, there has often been accompanying shortcomings. Availability, spatial resolution and restrictive vertical perspective have limited the usability of satellite-generated datasets.
Apart from their low cost, drones combined with image processing software can provide frequent surveys of rapidly changing areas without cloud coverage issues and also offer a much more reliable oblique perspective. All these advantages and the very high resolution output generated by software like Pix4Dmapper have placed UAVs in the spotlight of the disaster response community. During operational training in the field, participants worked alongside DJI, Pix4D, and the Community Disaster Management Committee (CDMC) of Panga — a village that had been badly damaged in the earthquake — to create a complete map of the area. Using Phantom 3 Advanced quadcopters and Pix4Dmapper, orthomosaics were produced overnight and the local community can now use them for the reconstruction process as well as create preparedness plans for future events. While this training had a humanitarian base, the goal was to further validate how drones and image analysis can be used in disaster situations.
Data collection
When the team arrived on the site, it coordinated with the local Community Disaster Management Committee (CDMC) to see which areas needed to be mapped the most. Permission to fly had already been obtained from the Civil Aviation Authority (CAAN).
Because the streets were littered with debris, powerlines and people around, the team had to climb on the rooftop of the highest surrounding building to ensure safe flying as well as to keep the drone in our line of sight.
The Pix4Dmapper Capture app was used to choose a grid mission that is optimal for most mapping. GPS was used to enable localization on a low resolution satellite map. Dragging the flight grid selection tool, which can scale to approximately 300×400 meters, depending on height, the team defined the area it wanted to map. Flight altitude, speed, image overlap percentage, and angle of the camera were all adjusted accordingly. The Phantom takes off automatically, acquiring images with a high overlap for a proper reconstruction in Pix4Dmapper. During the flight it was very important to keep eye contact with the drone at all times, so it could be quickly brought back in case of an emergency. In all, 9 small flights were taken, with a total flight time of 45 minutes, acquiring around 900 images at 3.4 cm resolution. Although a Phantom has up to 20 minutes of flight time, the flights on the field were short because of the high levels of airwave interference. Without this interference, the area covered could be much larger and by fewer flights, although for this project, most of the time expended was from climbing to the top of each building. Depending on the size of a project, different types of fixed wing or copter drones can provide different efficiencies.
The Phantom automatically came back to its starting point after the last image was taken for each flight. As soon as the drone landed, the team wanted to make sure that all pictures were taken properly, while they were still on site. They uploaded the images to the Capture Cloud service that computes a 2D and 3D preview within minutes after the images automatically upload on the phone itself, and viewed the orthomosaic on our phone’s browser. For processing, the images were transferred to the desktop.
Processing the data
Using Pix4Dmapper on the laptop, the team selected the 3D Maps template with default (WGS84) coordinate system. The software’s automatic processing comprises 3 main steps: The first step optimizes camera positions and analyzes image information, extracting keypoints and matching them across the images. The second step builds a 3D point cloud and model, and the third step generates the DSM and orthomosaic.
All of the images from different flights were processed together in one project, taking around 70 minutes on a MSI laptop with an i7 quad core and GTX 970M GPU. For projects where this doesn’t work (flights may have very different resolutions for example), flights can be processed separately and merged together by creating manual tie points between the images. The built in tools of Pix4Dmapper, like the raycloud and mosaic editor lent accuracy and quality to the project. The final results were a 3D point cloud, 3D model, and 2D map (orthomosaic) of the village of Panga. Instead of vertical or close to nadir imagery with meter resolution as one might get with satellite maps, the team now had data that is at centimeter resolution and provided a more comprehensive perspective with its oblique viewing angle.
