Research on Temperature Field Reconstruction Based on the Fusion of 3D Point Cloud and Infrared Image of the Tunnel Face

Imagine you are a detective trying to solve a mystery inside a dark, dusty cave. Usually, detectives use a flashlight to see shapes and a thermal camera to spot heat. But here's the problem: the flashlight shows you where a rock is, and the thermal camera shows you how hot it is, but they don't speak the same language. The thermal picture is flat like a sticker, while the cave is deep and three-dimensional. If you try to stick the flat sticker onto the 3D cave, it might end up on the wrong rock, or it might look all blocky and pixelated, like a low-resolution video game. This is a huge headache for engineers building tunnels underground. They need to know exactly where dangerous hot spots (which might mean a rock burst is coming) or cold spots (which might mean water is rushing in) are located in 3D space, not just on a flat screen. Without a clear map, they can't warn workers in time.

This paper is about a clever new way to glue those two different pictures together perfectly. The researchers built a system that takes a bunch of regular photos of a tunnel wall to build a 3D model (like a digital clay sculpture) and then paints the heat map onto it so the temperature belongs to the specific rock it came from. They didn't just guess; they tested it in real tunnels with dust and bad lighting. They found that their new method is much better at sticking the heat map to the 3D shape than the old ways, and it can spot tiny temperature changes as small as 0.1 degrees Celsius. It's like upgrading from a blurry, mismatched sticker to a high-definition, custom-fit thermal suit for the tunnel wall, helping engineers see hidden dangers before they become disasters.

The Problem: The Flat Map vs. The 3D World

Think of a tunnel face (the wall at the very front of a tunnel being dug) as a giant, rough canvas. Engineers want to know two things about this canvas: its shape (is there a crack? is it bulging?) and its temperature (is it getting hot like a fever, or cold like a draft?).

To get the shape, they used to need expensive laser scanners. To get the temperature, they used infrared cameras. But the infrared camera only sees a flat, 2D picture. If you look at a flat photo of a hot rock, you don't know if that rock is right in front of you or far away in the back. If you try to paste that flat photo onto a 3D model of the tunnel, it often ends up in the wrong place, or it looks "stair-stepped" and jagged, making it hard to tell exactly where the danger is.

The old ways of fixing this had big flaws. Some methods required setting up heavy, rigid equipment that couldn't move, which is useless if you are walking through a tunnel with a backpack. Others tried to use artificial intelligence to match the pictures automatically, but in a dusty, dark tunnel with boring-looking rocks, the AI got confused and made mistakes.

The Solution: A "Virtual Depth" Guide

The team, led by researchers from institutions like the Nanjing Hydraulic Research Institute and the University of South China, came up with a three-step magic trick to solve this without needing expensive lasers or rigid setups.

Step 1: Building the 3D Model from Regular Photos
Instead of using a laser scanner, they used a standard camera to take many photos of the tunnel wall from different angles. They used a smart software tool (based on something called COLMAP) to stitch these photos together into a dense 3D point cloud. Imagine taking hundreds of photos of a statue and using a computer to figure out exactly where every single point on the statue is in 3D space. They even figured out how to make this model the right size by measuring the known width of the tunnel, so the digital model matches the real world perfectly.

Step 2: The "Virtual Depth" Trick for Registration
This is the coolest part. The biggest challenge was lining up the flat thermal image with the 3D model. Usually, you have to manually pick matching points on both, which is hard and error-prone. The researchers invented a "Virtual Depth View." They took their 3D model and projected it onto a 2D screen, creating a fake "depth map" that looks like a grayscale image where the brightness tells you how far away the rock is.

Now, instead of trying to match a thermal image to a 3D cloud (which is hard), they match the thermal image to this new "depth map" (which is easy because they are both 2D images). It's like using a shadow puppet to help you line up a sticker. Because the depth map has clear edges and shapes, the computer can find matching points much faster and more accurately. They call this a "semi-automatic PnP registration," and it works so well that it doesn't need any physical markers or calibration boards stuck to the wall.

Step 3: Smoothing the Heat Map
Once the thermal image is lined up with the 3D model, they have to paint the temperature onto the 3D points. The old way was "nearest-neighbor," which just grabs the closest pixel's temperature. This creates a blocky, mosaic look, like a pixelated video game. The new method uses "bilinear interpolation." Imagine you are blending colors on a canvas; instead of just picking one color, you look at the four pixels around a spot and mix their temperatures together. This creates a smooth, natural gradient, eliminating the ugly "stair-step" artifacts.

What They Found: The Results

The researchers tested their new system in real tunnels, which are often full of dust, water vapor, and poor lighting. Here is what they discovered:

• Superior Alignment: Their new "Virtual Depth" method was incredibly accurate. It achieved a "reprojection error" of less than 2.0 pixels. In comparison, the traditional manual method had an error of 4.8 pixels, and even a popular AI method called SuperGlue had an error of 2.5 pixels. Their method also succeeded 98.6% of the time, beating the others significantly.
• Pinpoint Accuracy: When they checked the temperature against real thermometers (thermocouples) placed on the tunnel wall, their 3D reconstruction was incredibly close. The average error was about 0.10 °C, and the biggest mistake was only 0.18 °C. This is well within the safe limit of ±0.5 °C needed for engineering safety.
• Spotting the Tiny Details: They could detect temperature changes as small as 0.1 °C. This is crucial because a tiny hot spot might be the first sign of a rock burst, and a tiny cold spot might mean water is leaking in. Their system filtered out the noise from dust and bad lighting to find these weak signals.
• Speed and Practicality: The whole process, once the photos are taken, takes about 2 seconds on a normal portable computer. It doesn't need a supercomputer or a fixed setup; a worker can just walk in with a backpack, take photos, and get the result.

Why This Matters

The paper explicitly rules out the idea that you need expensive laser scanners or heavy, fixed equipment to do this job. They also showed that relying on deep learning (like SuperGlue) alone isn't enough for the messy, low-texture environment of a tunnel.

Instead, they proved that a smart combination of standard cameras, a "virtual depth" guide, and smooth math can create a high-precision 3D temperature map. This isn't just a theory; they tested it in multiple real tunnel projects. The result is a tool that can help engineers see hidden dangers in the rock before they cause accidents, keeping construction sites safer and more efficient. It turns a flat, confusing heat map into a clear, 3D warning system that anyone can trust.