News, Blog & Videos
January 16, 2018
Aerial photogrammetry is a method for mapping topography where multiple overlapping images are taken across an area and are used to create orthophotos and 3D spatial models of the topography. Recently a large number of micro (> 5kg) unmanned aerial vehicles (UAVs) have become available at a reasonable cost. These UAVs are fully autonomous; come equipped with a GNSS/IMU navigation system, and can be outfitted with variety of sensing devices. The most common sensing devices are consumer grade digital cameras. When combined these technologies make for a viable photogrammetry tool for producing orthophotos and digital surface models.
It is important to note that the limitations in size and power of these UAVs means they can be unstable in windy conditions, have limited flight endurance, and cannot operate effectively in precipitation. There are also many regulations surrounding commercial UAV operations that must be adhered to. The advantages of this system include ease of deployment and quick delivery of imagery with high spatial resolution in areas of limited extent.
Automatic generation of 3D spatial models is made possible by combining both photogrammetry and computer vision algorithms. Before a model is generated there first has to be a large number of images with a high degree of overlap. Using the image set a Structure-from-Motion (SfM) algorithm is able to estimate the camera’s parameters, position, and orientation while simultaneously reconstructing the scene geometry. The SfM software does this through the automatic identification of matching features in the image set and then performing a highly redundant bundle adjustment of these matching points. Points of known location can be introduced for better georeferencing accuracy.
The amount of matching points will determine how dense the resulting point cloud will be; the density of the point cloud directly affects the accuracy of the 3D scene reconstruction. The spatial accuracy of the individual matching points will also determine the quality of the point cloud. The number of matching points is based on the texture of the image objects and the image resolution. If image objects are homogenous or untextured such as water, snow, or asphalt the resulting point cloud may be sparse. Sparse point clouds produce a poorer representation of the elevation surface. Patchy snow cover may still produce an adequate surface model but in general it is better to avoid snow cover. In order to produce quality surface models requires the skill and knowledge to utilise this technology effectively.