Why overlap matters
Methods for capturing and processing data with low overlap have recently garnered some attention in the agricultural drone industry. The idea is that by reducing the amount by which images overlap, overall flight time is reduced, ultimately saving drone pilots time and money. While it’s true that this method does save time, the bottom line is that it presents an important tradeoff: it sacrifices accuracy, resolution, and valuable outputs to save an extra five minutes in the field.
That’s right — we found that flying with standard overlap (70%) only adds an additional five minutes to the flight plan, compared to flying the same field with low overlap (25%). ***
Here are five reasons to fly for five more minutes:
One vital output that can only be produced from standard-overlap data is a Digital Surface Model (DSM). A DSM is an astonishingly advantageous tool in any agronomist’s arsenal, primarily because of its use in evaluating surface properties and water flow. For example, in the image below, an agronomist used a combination of NDVI and a DSM to determine that a large area of his client’s potato field was suffering from overwatering. Problems like this are difficult to identify through vegetation indices alone, which tend to focus only on plant stress or disease.
DSMs are produced via an advanced photogrammetric process that compares and matches common points taken from different angles in each image to build a three-dimensional model of the captured area. The more matches there are, the more accurate the model and the fewer artifacts that appear in the resulting mosaic.
In contrast, low-overlap processing methods stitch images together similar to the way your mobile phone creates panoramas; with few common points between the images and a large amount of blending. This process cannot create any 3D outputs because there aren’t enough comparison points for an accurate model.
2. Sloped Terrain
When stitching images taken over varying terrain, high overlap is essential for accurate 3D modeling. Without it, the lack of common reference points in each image leads to artifacts in the resulting mosaic, such as misaligned objects and distorted distances.
With enough overlap, the output is a true orthomosaic which models the terrain accurately and can be used for precise measurements. Low overlap data will produce a quality result only if the terrain is perfectly flat and the camera was perfectly nadired (facing downward) during flight.
Plants absorb a lot of sunlight, but they also reflect much of it back into the sky and into the camera. Because neither the sun’s illumination nor the plant reflections are diffuse (evenly scattered in all directions), this causes bright spots and gradients in each of the captured images. The location of these bright spots depends on the combined orientations and angles of the sun and camera to the ground.
If there are multiple overlapping images that cover the same area, it’s possible to eliminate or significantly reduce gradients and sunspots by combining data from different images. In the example above, there are two images, each with a sun glare. The glare appears to be in a different location in each image due to the various angles involved. One simple way we can fix this is simply by overlapping the bottom image behind the top, and the top behind a third image, hiding the sun glare completely. However, this can only be done with significant overlap.
Furthermore, multiple overlapping images provide multiple measurements of the surface reflectance over different angles, which in turn provides the opportunity to create a more accurate reflectance map. This means that NDVI and other indices produced from the data will be more accurate.
High resolution data is beneficial for many specialty crops and is one of the main benefits of capturing data from a drone. With most standard photogrammetry solutions, the output resolution is typically the same as the input resolution; if 8 cm/px data is collected, the processed output will also be 8 cm/px.
Low-overlap solutions may not provide the full resolution of the data after processing. A common technique for reducing the impact of errors and artifacts in low overlap processing is to reduce the resolution of the output data. For example, if data is captured at 4 cm/px resolution, the processed output could be reduced to as much as 10 meters/px.
When evaluating various solutions, it’s important to ask if the processed output resolution is the same as the input resolution. Otherwise, valuable details might be lost in the process.
Low-overlap data requires different, often proprietary, software for processing. Usually, this means that users are restricted to specialty software and the associated subscription costs. With more data processing solutions on the market, data collected with standard overlap can be handled in a wide variety of solutions; users are not locked into any one system for processing and can choose the one that best suits their needs. There are even solutions available that can process standard overlap data in minutes, while still in the field. An important question to ask sensor providers is whether or not the data from their sensor can be processed reliably using multiple platforms.
The Bottom Line
While it’s true that flying with low overlap can save time, ultimately users get more from data collected with standard overlap, including DSMs, heightened accuracy, more reliable agricultural indexes, high-resolution data, and more processing options — all in the time it took to read this article.
Got five more minutes?
See how an agronomist used a DSM and other layers to pinpoint problems in a potato pivot. Or read about how high-resolution data helped a farmer identify chlorosis in plums. If you have a case study with RedEdge, Sequoia, or Atlas that you would like to showcase, please send us an email here.