Development of an Enhanced Aerial Photogrammetric Process for Precise 3D Model Reconstruction and validation by Volume Calculation

Imagine you are trying to build a perfect 3D statue of your favorite toy, but you only have a flat, two-dimensional camera. This is the world of photogrammetry, a branch of science that turns flat photos into 3D models. Think of it like a digital puzzle: if you take enough pictures of an object from different angles, a computer can figure out how deep the object is, just like your brain uses two eyes to judge distance. Usually, drones do this automatically, flying in perfect patterns to snap hundreds of photos. But what if the drone gets tired, or the battery dies, or the computer gets overwhelmed by too much data? That's where this research steps in, asking a simple but tricky question: Do we really need to take a million photos, or can we get the same perfect 3D model by taking fewer, smarter pictures?

The researchers in this study wanted to find the "sweet spot" for taking photos. They were looking for the perfect camera angles that would give them a super-accurate 3D model without wasting time or storage space. They tested different ways of looking at an object—straight down, from the side, or somewhere in between—to see which combination created the best digital copy. Why does this matter? Because in the real world, whether you are measuring a pile of dirt for a construction site or scanning a historic building, getting the volume (how much space it takes up) exactly right is crucial. If your math is off, you might order too much concrete or miss a structural flaw. This paper tries to find the most efficient way to get that math right.

The Great Photo Angle Hunt

The team at Garuda Aerospace and several engineering colleges in India decided to play a game of "guess the angle" to see how to build the best 3D model of a simple object: a paper box. They didn't just guess, though; they set up a very specific experiment. Imagine a flat circle on the ground, divided into twelve slices like a pizza. They placed the box in the center and stood at each slice to take a picture.

But here's the twist: they didn't just stand there. They tilted their cameras to look at the box from four different heights of view: 30°, 45°, 60°, and 90°.

• 90° is looking straight down, like a bird's-eye view.
• 0° (which they also tested) is looking straight at the side, like a human standing next to it.
• 30°, 45°, and 60° are the "oblique" angles, looking down at the object from a slant, like a drone hovering diagonally.

They started with a mobile phone camera held at a constant height of 5 feet. They took photos of the box from every angle, sometimes taking just one angle (like only 45°) and sometimes mixing them up (like taking photos at both 45° and 60°). After snapping the pictures, they fed them into a powerful computer program called Agisoft Metashape. This software acts like a digital sculptor, stitching the flat photos together to create a 3D model, a "Digital Elevation Model" (a map of heights), and a flat, perfect aerial map called an "Orthomosaic."

The Volume Test: Did the Math Work?

To see if their digital models were actually good, they had to play a game of "compare and check." They knew the exact size of the paper box because they measured it with a ruler. The real box had a volume of 0.00189 m³.

Then, they asked the computer to measure the volume of the 3D models it created from the different photo sets. They used another program, QGIS, to calculate the volume of the digital box and compared it to the real one. If the computer said the box was 0.00189 m³, they were perfect. If it said 0.00200 m³, they were a little off.

Here is what they found, and it was quite surprising:

• The "All-Over-the-Place" Mistake: When they tried to use every angle they had (0°, 30°, 45°, 60°, and 90°) all at once, the computer got confused. The error jumped to a massive +82.5%. It turns out, throwing too many different types of photos together can actually make the model worse, not better.
• The "Straight Down" Problem: Taking photos only from 90° (straight down) gave a decent result, but it was still off by -3.70%. It missed some of the side details.
• The "Side View" Struggle: Looking only from 0° (straight on) was a disaster for the software; it couldn't even match the points properly to build a model.
• The Winning Combo: The magic happened when they mixed 45° and 60° using the mobile phone. This specific combination of slanted angles created a 3D model that was incredibly close to the real box. The computer calculated the volume as 0.00190 m³, which was only +0.53% off from the real measurement. This was the most accurate result of the entire study.

They also tested a mix of 30°, 45°, and 60°, which was also good, with an error of +3.18%, but the 45° and 60° pair was the clear champion.

From Phone to Drone: The Real-World Test

Once they figured out the best angles with the phone, they wanted to make sure it worked with a real drone. They swapped the phone for a Garuda Aerospace drone and flew it over the object at a height of 30 feet. They used the same winning angles (45° and 60°) to take the pictures.

The result? The drone method was also highly accurate, with the error dropping to less than 2.76%. The drone's ability to fly smoothly and capture overlapping images from those specific slanted angles made the 3D model incredibly accurate. The researchers noted that this "non-autonomous" approach (where a human guides the drone to take specific photos rather than letting the drone fly a pre-set pattern blindly) saved time and reduced the amount of useless data collected.

The Bottom Line

This study suggests that you don't need to take a million photos or fly a drone in a complicated, automated grid to get a perfect 3D model. Instead, by manually guiding a camera (or drone) to snap photos at the 45° and 60° angles, you can create a 3D model that is highly accurate.

The paper highlights that the 45° and 60° combination yielded the highest accuracy, achieving an error of just +0.53% when processing the mobile phone images. When this optimized angle setup was applied to aerial photogrammetry using a drone, it also produced highly accurate volume estimates, with an error of less than 2.76%.

The paper concludes that these specific slanted angles are the "Goldilocks" zone—not too steep, not too flat, but just right. They provide enough detail of the top and sides of the object to help the computer build a solid, accurate 3D shape without getting confused by too much extra data. Whether you are using a phone or a drone, looking at your object from these two specific angles seems to be the secret to building the best digital twin possible.