TWM Contributor: Jeff Clay, PLS
Director of Reality Capture Services

The Wrong Equipment Can Lead To Costly Mistakes & Inaccurate Results

In recent years, using drones for topographic surveys has become increasingly popular across various industries. From construction and agriculture to environmental monitoring and urban planning, drones offer a cost-effective and efficient means of gathering data. However, the attempt to use low-budget drone-based equipment to achieve high-accuracy surveys presents significant drawbacks and disadvantages.

Two of the most common types of drones include a photogrammetry-based drone (overlapping photos are stitched together to create a 3D point cloud) or a LiDAR-based drone (send laser pulses to the ground to measure distance from drone to ground and produce 3D point cloud).

Photogrammetry-Based Drones:

Photogrammetry is notoriously difficult (despite what most dealers may try to sell it as) to get right, and has significant pitfalls – especially in regions with forests, prairie grasses, and other dense vegetation – such as:

  • Overgrown grass: If the site has overgrown grass, the resulting data that’s produced from the collection will represent the top of the grass. Obviously, this can lead to a bad/inaccurate surface and the subsequent grading plan will likely be compromised and/or cause the site to be unbalanced.
  • Tree canopies: Any tree coverage on the site and in areas that have tree canopies will result in no data or inaccurate data. Winter (leaf-off conditions) can help mitigate this issue to a degree, but most often, this dataset will show holes or “noisy” data in those areas (“noisy” data is data that doesn’t quite represent the actual ground surface).
  • Vegetation: Photogrammetry would be ideal in an arid environment, where there is hardly any vegetation to see through, though the homogenous scenery can cause issues with photogrammetric reconstruction as well.

The ideal type of work for photogrammetry is typically creating 2D-drawings (i.e. drawing linework/planimetrics). While TWM typically doesn’t often use strictly photogrammetry for topographic surveys due to a lot of these challenges and pitfalls, our geospatial group prefers to use LiDAR for wide-area topographic surveys.

Even when the field collection, photogrammetry processing, and post-processing (building a surface) are all correct, these pitfalls/disadvantages are the best-case scenario. Some common processing errors include:

  • The photogrammetric model having a “bowl” shape or some sort of distortion due to various issues, such as not enough overlap in imagery or due to not enough control on-site to constrain the photogrammetric model.
  • Imagery that’s too blurry, resulting in less accurate dataset/resulting surface.
  • Improper point filtering (i.e. incorrectly using the “lowest” points in any given area to use as the surface model). Every remote sensing technology (aerial LiDAR, static LiDAR, photogrammetry, etc.)  generates a point cloud that has some degree of “noise,” which is points that do not represent the actual surface of whatever object is being modeled (which, in our case is usually ground). Photogrammetry will often create points in the point cloud that are below the surface of the ground (I’ve seen shadows cause the point cloud to deviate downwards 2’ on a road). If the unfiltered point cloud is used to create a DTM, usually it will catch these “low points” and create the surface from the lowest point in a specified grid. Similarly, vegetation (bushes, trees, etc.) can cause the surface to go up and the ground will be “represented” on top of the bushes because filtering didn’t remove these as not-ground points.

LiDAR-Based Drones:

LiDAR can mitigate some of the potential issues or challenges of photogrammetry, such as the active laser that sends a pulse of energy to the ground and will often get “multiple returns,” which will usually provide a better opportunity at “seeing” ground through vegetation. LiDAR can sometimes provide data that’s more precise than photogrammetry, but it still has its own set of challenges:

  • There is a wide range of LiDAR sensors (cheap to expensive) and buyers typically get what’s paid for. A more expensive system is going to have better precision (less “noise” in the point cloud – i.e. ground profile looks thin/crisp/clean) and will be able to capture more data than photogrammetry in areas where there is vegetation coverage.
  • A cheaper LiDAR system (such as DJI L1/L2, or Rock Robotic R3/R3 Pro, etc.) will typically provide more data and sometimes slightly better data than photogrammetry, but the biggest upside is seeing through the vegetation. Often, the biggest issue with these cheaper systems is they are typically paired with low-grade IMUs, which measure the change in orientation and position of the drone across time less accurately than a high-grade IMU.

However, there are still other pitfalls to consider when using LiDAR-based drones, including:

  • A steep learning curve for correctly processing the trajectory (cloud solutions typically don’t do very well – many times it requires quite a few iterations, changing some processing settings, etc.).
  • A steeper entry price than a photogrammetry rig.
  • Data that’s sometimes more susceptible to points beneath ground surface (mostly due to shiny surfaces – cars, windows, etc.).
  • Data that still requires significant filtering to create a good surface model.

Regardless of the type of drone-based equipment that’s used, there are some major factors that will determine success, including:

  • Determining proper workflows for all aspects of processing: field collection, processing (pre- and post-processing), and deliverables generation.
    • To maintain full control over data and processing, we recommend avoiding cloud processing. Data could “look” good but not be good (i.e. good from far, but far from good). This is especially important for the post-processing (filtering) and deliverables generation. If you have no way to view, edit, manipulate, or check/QA the data and deliverables, it’s not a matter of if you have a project go bad, it’s a matter of when. TWM decidedly invests in multiple point cloud processing software subscriptions/maintenance each year to remain efficient and ensure that we’re delivering highly accurate data. We will typically use 2-4 different software for a typical photogrammetry project and 6-7 different software for a typical LiDAR project.
  • Following ASPRS Positional Accuracy Standards for Digital Geospatial Data: Having enough GCPs and check shots (and knowing the difference between the two) on the project site to have a statistically sound error report. TWM follows ASPRS Standards for all our projects.

What Went Wrong?
Here are a couple of project examples where TWM’s expertise was needed to resolve issues for our clients:

Example #1:
Our team received photogrammetry data from a local County of a road that was collected and processed by another company. TWM was not involved at this point. This survey was provided to our team to redesign the road. However, TWM needed to locate a few culverts that weren’t included in the original survey, and noticed they were FEET off vertically from the photogrammetry surface. This wasn’t a shift, the errors undulated up and down. Our team ended up driving this roadway with mobile LiDAR and processed a new survey.
The Issue: This project was a straight road, and the original photogrammetry survey had GCPs along the center of the road. No other GCPs were out near the extents of the project, which allowed for twisting of the model (think walking a tight rope). This twisting in the photogrammetric model allowed for variable errors, both horizontal, but more critically, vertical errors.

Poor example (above): drone LiDAR point cloud of curb profile from budget competitor

Ideal example (above): TWM’s drone LiDAR point cloud of curb profile

Example #2:
Data collection and processing/deliverables for this LiDAR project for a federal government agency were supposed to be done by a sub-contractor. The sub-contractor processed everything in the cloud without any software to view/edit/QA/manipulate the LiDAR data. The cloud report said errors were within 0.1’ vertically (sounds great!), but when our team reviewed the point cloud data against the ground control, the survey was off up to 0.5’ vertically.
The Issue: Some settings were being incorrectly set (such as allowing a full 180° view of LiDAR), which allowed for LiDAR points from the complete opposite side of the site  to be included when adjusting  to the control points. The points – when directly beneath where the LiDAR data is being collected – would have very tolerable errors but once that data is projected one side of the site to the opposite side of the site, the error is compounded and resulted in data to be shown below the ground surface. The cloud service then adjusted all the data to the “noise”, or inaccurately below the ground surface data. There was also some “noise” from cars being nearby and the shiny paint/windows cause points to appear below the surface as well. These were not filtered out after the settings were adjusted to narrow down to approximately 90° field of view. Narrowing down the field of view from 180° to 90° limits error propagation across large distances and typically will generate a more precise point cloud.

Both of these projects appeared to be of good quality on the surface (good from far, but far from good) but once our team dug in deeper, we recognized major issues that needed to be addressed before each project could move forward. And without proper QA procedures in place, the issues could have been pushed further down the design pipeline, causing a more costly issue either during design, or during construction.

The Value of Investing
While low-budget drone equipment may seem to offer appealing cost savings, its limitations and inaccuracies can significantly impact and compromise survey operations. It’s essential for surveyors to invest in high-quality drones that are equipped with advanced sensors and technologies. Without precise and reliable data, and the training required to successfully operate this equipment, the result will be data loss, project delays, and additional expenses. That’s why TWM always prioritizes superior quality and performance to mitigate risks, maximize efficiencies, and deliver on our promise of providing Exceptional Service. Nothing Less.

About TWM’s Geospatial Group
TWM separates itself from the competition by using state-of-the-art equipment for data collection. Depending on various aspects of your project, such as size, accuracy, and environmental conditions, TWM has the resources to collect highly detailed site conditions and capture data on almost any project. Our firm owns and operates reality capture tools including high-definition 3D laser scanners for precision data that’s ideal for tight or confined interior spaces, mobile LiDAR to collect data at highway speeds, and UAVs for aerial LiDAR or photogrammetry/imaging. By using this technology, TWM can help design professionals save time, increase revenue, minimize risk, and correct conflicts prior to construction.

Read more about our geospatial engineering services. Visit our Projects page to explore some of TWM’s aerial LiDAR projects.