Design of a Smartphone Indoor Positioning Dynamic Ground Truth Reference System Using Robust Visual Encoded Targets

Xuan Liao, Ruizhi Chen, Ming Li, Bingxuan Guo, Xiaoji Niu, Weilong Zhang2019 · DOI

TL;DR

This paper introduces a novel, low-cost system for generating accurate, real-time "ground truth" for smartphone indoor positioning. By using robust, color-encoded visual targets, the system can dynamically determine a smartphone's precise pose, overcoming the limitations of existing static or expensive methods. This enables more reliable evaluation of new indoor positioning technologies.

What Problem Does This Paper Address?

Accurately measuring the performance of indoor positioning systems is crucial for their development. However, obtaining a precise, real-time "ground truth" (the actual, known position of a device) for smartphones indoors is challenging. Existing methods like high-precision robots or laser SLAM systems are expensive, cumbersome, or not suitable for dynamic, real-world scenarios. This gap makes it difficult to reliably evaluate and compare different smartphone indoor positioning techniques. The paper aims to solve this by creating an accessible and accurate dynamic ground truth reference system.

Methods

The core of the proposed system is a set of "robust visual scatter-encoded targets." These targets are designed with a specific arrangement of colored circular points (red for base points, and white, green, and black for identification points) on a black background. This design enhances recognition rates and provides a stable reference.

The process involves:

  • Target Design: Creating unique color scatter-encoded patterns with a high encoding capacity (177,147 possibilities) and invariance to scaling, rotation, and translation.
  • Base Point Extraction: Using smartphone images, the system first extracts the red "base points" of the encoded targets. This involves edge detection (Canny operator) followed by ellipse fitting to accurately determine the center of these circular points, even when they appear as ellipses due to perspective.
  • Local Coordinate System Setup: A local coordinate system is established based on the relative positions of these base points, leveraging the principle of cross-ratio invariance to ensure robustness against image distortions.
  • Decoding Encoded Targets: The system decodes the color information of the identification points (white, green, black) to assign a unique ternary code to each target pattern. This code is then converted to a decimal value, effectively identifying the target.
  • Pose Solution: Using the decoded target's known 3D coordinates and the corresponding 2D image coordinates, the system employs space resection and least squares methods to calculate the smartphone camera's precise pose (position and orientation) in real-time.

Datasets

The paper does not explicitly mention pre-existing datasets for training. Instead, it describes the creation of a custom dataset.

  • Custom Dataset: The researchers collected 54 images of the designed color scatter-encoded target patterns in an experimental room (10m x 9m). These images were used to build the "Encoded Targets Library," which stores the decoded values and corresponding 3D coordinates of the targets. This library serves as the ground truth for subsequent positioning experiments.

Key Insights and Ideas

  • Color + Geometry for Robustness: Combining color information with geometric structure in the encoded targets significantly improves recognition rates and fault tolerance, making the system less susceptible to lighting variations and image distortions.
  • Cross-Ratio Invariance for Stability: Utilizing the cross-ratio invariance principle for base point selection provides a robust method for establishing the local coordinate system, even with imperfect image data.
  • Dynamic Ground Truth: The system's ability to dynamically determine the smartphone's pose in real-time is a significant advantage over static ground truth methods, allowing for more realistic performance evaluations.
  • Low-Cost and Accessible: The use of standard smartphone cameras and easily printable encoded targets makes this a cost-effective solution for generating ground truth, democratizing the evaluation of indoor positioning technologies.

Key Results

  • High Recognition Rate: The color scatter-encoded targets achieved a 100% recognition rate in the experiments.
  • Centimeter-Level Accuracy: The proposed system demonstrated centimeter-level positioning accuracy, which is significantly better than typical indoor positioning methods like Wi-Fi or Bluetooth.
  • Robustness: The system showed good robustness to different illumination conditions and camera angles, and it could correctly reject incorrectly structured targets.
  • Smartphone Agnosticism: The method proved applicable to different smartphone models (Samsung Galaxy S8 and Huawei P10) with consistent accuracy.
  • RMSE: The Root Mean Square Error (RMSE) for overall coordinates was around 0.08-0.09 m, indicating high precision.

Limitations and Future Work

  • Capturing Distance Influence: The paper acknowledges that capturing distance can influence positioning results, although the high recognition rate of the encoded patterns mitigates this effect within a certain range.
  • Camera Surface Material: It was noted as difficult to measure the camera pose directly on the smartphone due to its glass material, necessitating the use of a ring crosshair for automated tracking.
  • Future Work: While not explicitly detailed, the authors suggest this system is a valuable ground truth reference for other smartphone indoor positioning technologies, implying its use in further research and development. The robustness and accuracy demonstrated suggest potential for integration into more complex indoor navigation systems.

Summary updated

BibTeX
@misc{liao2019design,
  author = {Xuan Liao and Ruizhi Chen and Ming Li and Bingxuan Guo and Xiaoji Niu and Weilong Zhang},
  title = {{Design of a Smartphone Indoor Positioning Dynamic Ground Truth Reference System Using Robust Visual Encoded Targets}},
  year = {2019},
  doi = {10.3390/s19051261},
}
Figure