💻 computer science

R-C-P Method: An Autonomous Volume Calculation Method Using Image Processing and Machine Vision

This paper introduces the R-C-P method, an autonomous volume calculation technique that utilizes image processing and edge detection with multiple 2D cameras to determine real-time dimensions and surface areas of objects as a cost-effective alternative to 3D depth sensors like LiDAR.

Original authors: MA Muktadir, Sydney Parker, Sun Yi

Published 2026-05-14

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: MA Muktadir, Sydney Parker, Sun Yi

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to guess the size of a box sitting on a table, but you don't have a tape measure, and you can't touch it. Usually, to get the exact size of a 3D object, you'd need a fancy, expensive "3D scanner" (like a high-tech laser or a depth-sensing camera) that acts like a super-accurate eye. But these scanners are heavy, eat up a lot of battery power, and are hard to fit into tight spaces.

This paper introduces a clever, low-cost trick called the R-C-P Method (Row-Column-Pixel). Think of it as using two regular, cheap 2D cameras (like the ones on your phone) to "guess" the volume of a box by looking at it from two sides, rather than using a super-expensive 3D scanner.

Here is how the method works, broken down into simple steps:

1. The Setup: Two Eyes, One Box

The researchers set up two cameras on opposite sides of a rectangular box. Instead of trying to see the "depth" (how far back the box goes) directly, they just take a flat picture of the box's face from each side.

2. Turning the Picture into a "Connect-the-Dots" Game

Once the cameras take the picture, the computer does some magic image processing:

Grayscale: It turns the colorful photo into black and white.
Edge Detection: It highlights the outline of the box, turning the picture into a sketch where the edges are white lines and the inside is black.
The "Fill-In" Trick (The R-C-P Part): This is the most creative part. Sometimes, the camera misses a tiny piece of the edge because of a shadow or a glitch. The R-C-P method acts like a smart "connect-the-dots" game.
- It looks at every row of pixels. If it sees a white dot on the left and a white dot on the right, it assumes the space between them is part of the box and fills it in with white.
- It does the same thing for every column (up and down).
- The Result: Even if the outline was broken or messy, the computer "fills in the gaps" to create a solid, perfect white rectangle. This gives them a clean shape to measure.

3. From Pixels to Real Life

Now the computer knows how many white pixels make up the box's face. But pixels aren't inches or centimeters.

The researchers figured out a "conversion rate": they know that a specific number of pixels equals a specific millimeter length (about 0.264 mm per pixel unit in their setup).
By counting the pixels and multiplying by this rate, they can calculate the width and height of the box as seen by the camera.

4. The "Distance" Problem

Here is the catch: If you move the box closer to the camera, it looks bigger. If you move it away, it looks smaller. The raw pixel count changes based on distance.

To fix this, the researchers ran many experiments moving the box closer and further away.
They created a set of mathematical formulas (equations) that act like a translator. If you tell the computer, "The box is 500mm away," the formula automatically adjusts the pixel count to tell you the real size of the object, regardless of how far away it is.

5. The Final Calculation

With two cameras, they get the width from one side and the height from the other. They plug these numbers into a volume formula (Width × Height × Depth) to estimate the total volume of the box.

What Did They Find?

It Works: The method successfully calculated the volume of rectangular boxes using only two standard cameras.
It's Forgiving: Even if the camera couldn't see the edge perfectly (due to a blurry spot or a gap), the "fill-in" trick fixed the shape so the calculation still worked.
The Accuracy: The results were pretty good, usually within a 4% to 10% margin of error for width, and slightly higher for height. The errors were a bit larger when the object was very close to the camera, but the math helped correct most of it.

The Bottom Line

The paper claims that you don't need a $1,000 3D depth sensor to measure the volume of a rectangular object. By using two cheap cameras, a bit of computer vision to "fill in the gaps," and some math to account for distance, you can get a surprisingly accurate estimate of an object's size in real-time. This is especially useful for robots or systems where saving money, space, and battery power is more important than having perfect, millimeter-precise measurements.

Technical Summary: R-C-P Method for Autonomous Volume Calculation

Problem Statement
While 3D depth sensors (e.g., LiDAR, Radar) are effective for situational awareness in autonomous systems, they introduce significant drawbacks, including high power consumption, large data volumes, complex cabling, and strict mechanical/thermal requirements. These factors render them unsuitable for specific operational environments, such as space. Conversely, 2D cameras offer a cost-effective, low-power, and space-efficient alternative. However, 2D cameras inherently lack depth information, making real-time volumetric estimation challenging without prior knowledge of camera pose or object categories. Existing methods often require multiple views (minimum of five), complex 3D scanning equipment, or specific texture patterns, which may not be feasible in all scenarios.

Methodology
The study proposes the Row-Column-Pixel (R-C-P) method, an autonomous volume calculation technique utilizing two fixed 2D cameras positioned on opposing sides of a rectangular object. The process involves the following steps:

Image Acquisition and Pre-processing: Two cameras capture RGB images of the object's surfaces. These images are converted to grayscale and then processed using Canny Edge Detection to identify object boundaries.
The R-C-P Algorithm:
- Step 1 (Row Direction): The algorithm scans each row of the binary edge image. If a white pixel (edge) is detected, the algorithm searches for a parallel white pixel in the same row. If found, all black pixels between these two points are converted to white, effectively "filling" the row.
- Step 2 (Column Direction): The algorithm performs a similar scan in the column direction. If a white pixel is found, it searches for a parallel white pixel in the same column, converting intervening black pixels to white.
- Result: This two-step filling process generates a solid binary mask representing the surface area, even if the original edge detection was discontinuous or incomplete.
Dimension Calculation:
- The total number of white pixels in the output corresponds to a physical dimension based on a calibration factor ( $d = 0.264$ mm).
- To derive actual physical dimensions (width and height) from pixel counts, the study establishes a relationship between the camera-to-object distance ( $x$ ) and the ratio of actual dimensions to R-C-P calculated dimensions.
- Linear regression equations (Equations 2–7) were derived to calculate correction factors ( $r_1$ for width, $r_2$ for height) based on the distance $x$ and the raw pixel measurements.
Volume Estimation: The final volume is calculated using the corrected dimensions from both cameras:
$V = W_1 \times \frac{(H_1 + H_2)}{2} \times W_2$
Where $W$ and $H$ represent the corrected width and height derived from the respective camera views.

Key Results
The method was validated using three rectangular objects at varying distances (395 mm to 645 mm).

Edge Robustness: The R-C-P method successfully reconstructed full surface areas even when Canny edge detection produced incomplete or discontinuous edges, demonstrating resilience to noise and detection gaps.
Accuracy:
- Width (Row): Excluding the closest distance (395 mm), errors in width calculation were typically below 4%, with a maximum of 10%.
- Height (Column): Excluding the 395 mm distance, height errors were mostly below 7%, reaching up to 17% in specific cases.
- The study notes that the 395 mm distance yielded higher errors, likely due to camera focus range limitations.
Correlation: Strong linear relationships were established between the distance from the camera and the ratio of actual to calculated dimensions, allowing for the derivation of the correction equations.

Significance and Claims
The paper claims that the R-C-P method offers a unified solution for shape reconstruction and volume estimation using only two fixed 2D cameras, requiring fewer views (two) than many existing 3D reconstruction methods (which often require five or more).

Cost and Complexity: It provides a viable alternative to expensive 3D depth sensors, making real-time volumetric monitoring feasible for budget-constrained or space-limited applications.
Autonomy: The system operates without prior knowledge of the object's category or the camera's precise pose, relying instead on the geometric relationship between the two fixed views.
Limitations: The authors acknowledge that the method currently relies on rectangular objects and is sensitive to environmental factors like lighting and camera calibration. The error rates, while acceptable for monitoring changes, indicate that the method is an approximation rather than a precision metrology tool in its current form.

Conclusion
The study demonstrates that 2D machine vision, enhanced by the R-C-P algorithm and distance-based correction equations, can effectively estimate the volume of rectangular objects in real-time. This approach eliminates the need for 3D sensors while maintaining sufficient accuracy for applications requiring volumetric data, such as object surveillance and defect detection, provided that environmental variables are controlled and calibration is refined.