Common Pitfalls of Point Clouds in Drone Surveying - and How to Avoid Them
Working in point clouds is difficult for three main reasons. This article will help you avoid these common pitfalls and provide a few tips for successfully integrating point clouds into your drone surveying workflow.
Ever since the introduction of LiDAR, point clouds have been a hot topic of discussion and debate among surveyors. They provide a rich and highly accurate source of data, but their complexity and size cause many firms to struggle when integrating point clouds into their data processing workflow.
Because of the rise in popularity of drone-based photogrammetry, the point cloud discussion has gained momentum. Point clouds are one of the core outputs of photogrammetry; some firms are very familiar with laser scanning technologies and successfully use point cloud data, but many do struggle and decide not to use point clouds at all, while others wind up using point clouds in a way that takes way too much time, costs too much money, and actually leads to a lower quality deliverable. We want to help you avoid the latter.
Although point clouds from drones are an incredibly rich and valuable source of data, they are often misused. When used properly, point clouds can serve an important role in a well-developed drone surveying system. In this article, we will examine some common pitfalls of point clouds, and discuss how to avoid them.
The Benefits and Challenges of Point Clouds
Point clouds are the richest, most complete, and most accurate source of data that comes out of drone photogrammetry. If the photogrammetry step has been processed well, then any bit of information contained in the photos will be represented as accurately as possible in the point cloud. So, if the data is so good, then why not use point clouds all the time and for every project?
The problem is that working in point clouds takes a lot of time, and if your drone program isn’t saving time and money, then it isn't working. So, if it is taking too long, then it probably makes sense to default to alternative methods rather than spending a week parsing through a massive point cloud file.
There are three key challenges that make working in point clouds difficult:
1) Too Much Data
The benefit of point clouds is that they capture absolutely everything about a project site. But that is also their downfall. When surveying a project site, surveyors almost never need to know the location of every leaf and every branch on every tree. They don’t need to know how many bumps there are on a manhole cover. They don’t need to know the exact dimensions of the roof rack strapped to the car parked on the edge of the property. And yet all that data is there. In order to get a clean deliverable, all of that data would have to be properly cleaned out, which can be very time-consuming.
2) Unclassified Points
Part of what makes working in point clouds so time-consuming is that, by default, drone point clouds typically aren’t classified in any meaningful way. With the advancement of machine learning and artificial intelligence, automated programs are getting better at automatically classifying points as surface, vegetation, objects, buildings, etc., but from first-hand experience, these algorithms still have quite a ways to go before they can become useful. So, for now, it is up to the surveyor to determine what is what in a point cloud and classify the data in the way they need it, whether that be curb lines, topographic features, buildings, or vegetation.
3) Difficult to Manage
Lastly, point clouds are very difficult to manage. The large file size, often over 10GB, can be difficult to transfer, and will usually crash familiar programs like Civil 3D without extensive modification. So instead, specialized point cloud software is used to extract data from the point cloud. This software can be extremely expensive and can still be difficult to use for all but the most experienced engineers and design technicians. Put simply, working in a point cloud takes a lot of the right types of hardware, software, and experience.
Tips for Working in Point Clouds
Working in a point cloud is not only possible but if done correctly, can be a valuable addition to an aerial surveying program. Here are a few tips we have found over the years that help when you want to work in point clouds.
Extract Selected Data
When working in a point cloud, you should go in knowing what data you need to extract, and then extract only that. A common mistake is to try and get everything, and it winds up being a time killer. Being brutally selective about what data you are trying to extract will help to save a huge amount of time.
Don’t force a square peg into a round hole
Or rather, don’t try to work with a point cloud in a program that isn’t designed for it. A common mistake that we see is when people delete 90% or more of points to make the point cloud smaller so it will work in familiar applications like Civil 3D. This is a process we call “Dumb Decimation”, and it results in losing 90% of the accuracy! Important features, like utilities, fire hydrants, signage, and curbs often disappear entirely in this method, and what is left is of lower quality than it was before. When working in point clouds, use a program designed to work with point clouds, and only ever use the full resolution file.
Know the limitations of photogrammetry-based point clouds
Photogrammetry based point clouds may be the richest source of data from photogrammetry, but that doesn’t make it perfect. While LiDAR often can penetrate at least some vegetation, the point clouds that come out of photogrammetry still only map the tops of vegetation. This means the top of tall grass, and the tops of trees when present. No amount of point cloud editing will create points where they don’t exist. As long as you are aware of these limitations going in, you should be fine.
If it isn’t working, use something else
Point clouds are not the solution to every problem. If you find yourself spending too much time working in a point cloud, and you just aren’t getting the data you need, don’t be afraid to collect more data in the field or on the ground, and use that to supplement your drone data. Or, don’t be afraid to abandon point clouds entirely and use an alternative method of extracting data from drone photogrammetry.
Alternatives to Point Clouds
If point clouds aren’t working for you, don’t give up hope! There is a huge amount of value to be had from drone photogrammetry, and not all of it comes from the point cloud. Here are a few other methods that can be successful in certain instances.
One of the simplest ways to get accurate data out of drone photogrammetry outputs is drafting data directly onto the orthophoto. This has the benefit of being extremely easy, very intuitive, and very accurate for XY coordinates at least. The obvious, massive downside to this is that the resulting data is not going to have any elevation data associated with it, so the applications are clearly limited.
2.5D modeling comes from combining a raster DSM file with the orthophoto, both of which are typically standard outputs of any decent photogrammetry software. By combining these into a 2.5D surface model, it is a lot easier to extract individual points and polylines from a surface than it is to work in a point cloud. The downside is that, because the model is only 2.5D, it is impossible to “look under” any eaves, trees, overhangs, or equipment to the surface below. But while the point cloud typically has richer data than the 2.5D model, often the time savings more than makes up for the slight loss in data quality.
Hire an Outside Expert
Many people make a career out of working in 3D data. That, combined with specialized computer hardware and software, means that specialists are often able to extract data from a point cloud faster, cheaper, and more accurately than others. Companies like Aerotas specialize in this type of work and can help you get the best data in the most efficient manner possible.
Using Point Clouds for Drone Surveying
Point clouds are a mixed blessing for many people. Their richness and accuracy have enabled surveyors to collect and analyze data that was never before possible. But they have also turned into a massive time sink, costing hundreds of hours that could have been spent better on other projects. The best way to use point clouds in drone surveying is to make sure that you use them selectively. Point clouds are not a magic bullet that solves all problems, so only use them when you know they can provide value. Also, be sure to use point clouds as part of a mixed workflow. Combining point cloud data with orthophoto data, 2.5D data, and good old-fashioned field surveying typically leads to the best end result.