Visible Light Positioning With Lamé Curve LEDs: A Generic Approach for Camera Pose Estimation

This paper proposes a generic Visible Light Positioning (VLP) algorithm called LC-VLP that utilizes Lamé curves as a unified representation for diverse LED shapes, enabling accurate camera pose estimation through a correspondence-free initialization and nonlinear optimization, which achieves superior performance over state-of-the-art methods with sub-4 cm average position accuracy.

The Gist

Imagine you are walking through a smart building with a smartphone in your hand. You want your phone to know exactly where you are and which way you are facing, but GPS doesn't work well inside. Usually, the building would need to be filled with hundreds of tiny, identical lights to help your phone triangulate its position.

But what if the building has a mix of old round lights, new square panels, and diamond-shaped fixtures? Most current "smart" positioning systems would get confused and fail because they are programmed to recognize only one specific shape.

This paper introduces a clever new solution called LC-VLP (Lamé Curve Visible Light Positioning). Think of it as a universal translator for light shapes. Here is how it works, explained simply:

1. The "Universal Shape" Trick (The Lamé Curve)

Imagine you have a piece of clay. You can squish it into a circle, stretch it into a rectangle, or pinch it into a diamond. In math, there is a special formula called a Lamé Curve that can describe all of these shapes just by changing a few numbers.

Instead of teaching the camera, "I know circles" or "I know rectangles," this new system teaches the camera: "I know the Lamé Curve."

• If the light is round, the system adjusts the numbers to look like a circle.
• If the light is square, it adjusts the numbers to look like a square.
• If it's a weird diamond, it adjusts again.

This means the system doesn't care if the ceiling is a chaotic mix of different light shapes. It sees them all as variations of the same mathematical family.

2. The "Shadow Game" (Back-Projection)

How does the camera figure out where it is?
Imagine you are holding a flashlight and shining it at a wall with a weirdly shaped hole in it. The shadow on the wall tells you something about your position relative to the hole.

In this system, the camera looks at the LED lights on the ceiling. Instead of trying to guess the 3D shape from the 2D image (which is like trying to guess the shape of a cloud just by looking at it), the system does the reverse. It takes the 2D image of the light and mathematically "projects" it back up to the ceiling.

It asks: "If this image on my screen came from a Lamé Curve on the ceiling, where exactly must I be standing for the angles to match?"

3. The "Guess and Check" (Optimization)

The system starts with a rough guess of where it is. Then, it plays a game of "Hot and Cold."

• It calculates the distance between the projected image and the actual mathematical shape of the light.
• If the distance is big, it knows it's "cold" (wrong guess).
• It tweaks its position slightly and tries again.
• It does this thousands of times in a split second, refining its guess until the math fits perfectly. This is called Nonlinear Least-Squares Optimization.

4. The "Magic Start" (FreePnP)

Usually, to start this "guess and check" game, you need to know exactly where you are to begin with. But this paper introduces a trick called FreePnP.

Imagine trying to solve a puzzle without the picture on the box. The system looks at the lights and says, "I don't know the exact corners, but I know the center of the light and the edge of the light are in a straight line." By using this simple rule of geometry, it can create "virtual" reference points out of thin air. This gives the system a good enough starting point to begin its "guess and check" refinement, even if it has never seen this specific room before.

Why is this a big deal?

• Flexibility: You can mix and match any light shapes in a room (round, square, oval, diamond), and the system still works.
• Accuracy: In tests, it was much more accurate than previous methods, getting within 4 centimeters (about 1.5 inches) of the true location.
• No Extra Hardware: It uses the camera you already have in your phone and the lights already on the ceiling. No need for expensive new sensors.

In short: This paper teaches a camera to speak the "language of shapes" so it can find its way in a room, no matter what kind of lights are hanging from the ceiling. It turns a chaotic mix of light fixtures into a precise GPS for the indoors.

Technical Summary

1. Problem Statement

Camera-based Visible Light Positioning (VLP) is a promising technology for indoor 6-DoF (Degrees of Freedom) camera pose estimation (CPE). However, existing state-of-the-art (SoTA) methods face significant limitations:

• Shape Specificity: Current methods are typically restricted to a single LED geometry (e.g., only circular or only rectangular). They fail in heterogeneous environments where different LED shapes (circles, rectangles, rhombuses, ellipses) coexist.
• Dependency on Auxiliary Sensors: Many methods rely on additional sensors like IMUs or magnetometers to resolve ambiguities, increasing hardware cost and complexity.
• Requirement for Pre-calibrated Points: Standard Perspective-n-Points (PnP) algorithms require $\ge 4$ known 3D-2D correspondences (Reference Points), which limits deployment flexibility.
• Sensitivity to Noise: Purely geometric solutions often lack robustness against image noise because they do not utilize global iterative optimization.

The core challenge is to develop a generic VLP algorithm that can handle arbitrary LED shapes without auxiliary sensors or extensive pre-calibration while maintaining high accuracy.

2. Methodology

The authors propose LC-VLP, a unified framework based on Lamé curves (superellipses). The methodology consists of three main stages:

A. Unified Modeling via Lamé Curves

Instead of treating different LED shapes as distinct problems, the paper models all common LED geometries (circles, rectangles, rhombuses, ellipses) using the Lamé curve equation:
$$\left| \frac{x - x_0}{a} \right|^\gamma + \left| \frac{y - y_0}{b} \right|^\gamma = 1$$

• $\gamma = 1$: Rhombus (Square if $a=b$).
• $\gamma = 2$: Ellipse (Circle if $a=b$).
• $\gamma \to \infty$: Rectangle (Square if $a=b$).
  An offline LED database is constructed to store the parameters ($x_0, y_0, a, b, \gamma, \phi$) for each LED, allowing the system to switch seamlessly between shapes without algorithm changes.

B. Correspondence-Free Initialization (FreePnP)

To avoid the need for pre-calibrated 3D-2D reference points, the authors developed FreePnP, a novel initialization algorithm:

• Collinearity Invariance Theorem: The paper proves that if three points are collinear in the 3D world, their projections are collinear on the 2D image plane.
• Virtual Correspondence Construction: By leveraging the geometric relationship between the centers of two visible LEDs and their projected contours, the algorithm constructs "virtual" 3D-2D correspondences.
• Sampling Strategy: Assuming uniform distribution of contour points, the algorithm samples points along the LED boundaries to generate a sufficient set of approximate correspondences.
• Result: A standard PnP solver (e.g., OPnP) is applied to these virtual correspondences to provide a reliable initial estimate of the camera pose ($R, t$).

C. Nonlinear Least-Squares (NLLS) Refinement

To achieve high precision, the initial estimate is refined using a back-projection strategy:

• Back-Projection: Instead of projecting 3D curves to 2D (which causes complex distortions), 2D image points are back-projected onto the 3D ceiling plane.
• Optimization Objective: The algorithm formulates an NLLS problem to minimize the algebraic distance between the back-projected points and the theoretical Lamé curve equations of the LEDs.
• Constraints: The optimization includes constraints for the camera's feasible region and, if available, pre-calibrated reference points to eliminate ambiguity.
• Solver: The problem is solved iteratively using Sequential Quadratic Programming (SQP).

3. Key Contributions

1. Generic Framework: First VLP approach to unify the modeling of diverse LED shapes (circular, rectangular, rhombic, elliptical) under a single Lamé curve parameterization.
2. FreePnP Algorithm: A novel initialization method that achieves coarse 6-DoF pose estimation without requiring pre-calibrated 3D-2D reference points, relying instead on geometric invariance and contour sampling.
3. Robust Optimization: A back-projection based NLLS refinement strategy that utilizes the full contour information of LEDs rather than just centroids or corners, significantly improving robustness against noise.
4. Comprehensive Validation: Extensive simulations and a physical prototype experiment verifying performance in both homogeneous and heterogeneous LED scenarios.

4. Results

Simulation Results

• Performance Gain: LC-VLP outperforms SoTA methods (V-PCA for circles, VLC-PnP for rectangles) in both homogeneous scenarios.
  • Position Error: Reduced by >40% compared to baselines.
  • Rotation Error: Reduced by >25% compared to baselines.
• Heterogeneous Scenarios: In mixed-shape environments (Scenario D), LC-VLP achieved a Mean Position Error (MPE) of 2.85 cm and Mean Rotation Error (MRE) of 0.35°, showing a 53% improvement over the best baseline (OPnP).
• Robustness: The algorithm demonstrated superior stability against image noise and varying LED sizes compared to geometric-only methods.

Experimental Results

• Prototype: A system was built using a realme Q2i smartphone camera and custom LED transmitters (circular and square) in a $2.4m \times 2.4m \times 2.4m$ room.
• Accuracy: The system achieved an average position accuracy of less than 4 cm (MPE = 3.94 cm) and an MRE of 3.80° across various camera heights and tilt angles.
• Stability: Performance remained consistent even when the camera was tilted by 30°, demonstrating the algorithm's robustness to orientation changes.

5. Significance

This paper addresses a critical gap in indoor positioning: the lack of flexibility in handling real-world, heterogeneous lighting infrastructures. By introducing Lamé curves as a universal shape descriptor and developing a correspondence-free initialization method, LC-VLP:

• Reduces Deployment Costs: Eliminates the need for auxiliary sensors and extensive pre-calibration of reference points.
• Increases Applicability: Enables VLP in complex environments (e.g., offices with mixed downlights and panel lights) where previous shape-specific algorithms would fail.
• Enhances Precision: Demonstrates that utilizing full contour geometry via nonlinear optimization yields significantly higher accuracy than traditional geometric approaches.

The proposed LC-VLP algorithm represents a significant step toward practical, high-accuracy, and low-cost indoor positioning systems for robotics, IoT, and smart building applications.