FILNet: Fast Image-Based Indoor Localization Using an Anchor Control Network

Sikang Liu, Zhao Huang, Jiafeng Li, Anna Li, Xingru HuangSensors 2023 · DOI

TL;DR

This paper introduces FILNet, a novel image-based indoor localization system that significantly improves accuracy and speed by using an "anchor control network" and an affine invariance enhancement technique. FILNet achieves faster feature matching and more robust localization, even under varying viewing angles, by intelligently indexing and filtering image features against a curated database of anchor points.

What Problem Does This Paper Address?

Indoor localization is a critical technology for applications ranging from building management to emergency response. While various methods exist (Bluetooth, WiFi, geomagnetic), image-based localization offers high accuracy and leverages ubiquitous smartphone cameras. However, existing image-based methods face challenges:

  • Privacy Concerns: Building detailed image databases can raise privacy issues.
  • Computational Cost: Feature matching can be slow and resource-intensive.
  • Robustness to Viewpoint Changes: Image perspective transformations (due to camera angle, distance, etc.) can severely degrade matching accuracy.
  • Efficiency: Existing methods often require expensive hardware or suffer from low efficiency.

This paper aims to address these limitations by proposing a fast, accurate, and robust image-based indoor localization system that minimizes privacy concerns and computational overhead.

Methods

FILNet operates in two main stages: offline database construction and online localization.

Offline Stage: 1. Anchor Control Network: Instead of a general scene database, FILNet uses "anchors" (e.g., murals) deployed in the environment. A subset of these are designated "control anchors" whose precise 3D coordinates are measured using a total station. 2. Visual-Inertial Odometry (VIO) with ARcore: For the remaining anchors, their 3D coordinates are inferred using VIO data from Google ARcore, leveraging the known control anchors. This reduces the need for manual measurement of every anchor. 3. Affine Invariance Enhancement: To combat viewpoint changes, an algorithm inspired by ASIFT is employed. It generates multi-angle features for each anchor point by screening and supplementing features from different perspectives. This makes the feature descriptors more robust to rotation, scale, and perspective transformations. 4. Feature Fingerprint Database: The enhanced, multi-angle features of the anchors are stored in a database. SURF features are chosen for their balance of speed and robustness.

Online Stage: 1. Fast Spatial Indexing: To speed up matching, FILNet uses a "fast spatial indexing" algorithm. It first identifies "active anchors" (those likely to be in the current view) and then performs feature matching only against features near these active anchors, significantly reducing the search space. 2. Homography Matrix Filtering: To further improve accuracy and remove false matches, a homography matrix is used. This verifies the geometric consistency of matched feature points, filtering out incorrect correspondences and improving the reliability of the final pose estimation.

Datasets

The experiments were conducted in a real-world indoor environment: the second floor of the THEOS remote sensing satellite ground station at Wuhan University. This environment included a floor plan of approximately 50m x 15m, with 8 visual control anchors and 14 indoor existing control points. For testing, 326 images were collected around 8 control points with varying view angles and distances. No new benchmark datasets were introduced; rather, existing real-world conditions were used for evaluation.

Key Insights and Ideas

  • Anchor Control Network for Privacy & Efficiency: Using pre-defined anchors (like murals) instead of general scene images significantly mitigates privacy concerns and allows for a more structured and efficient database.
  • VIO for Scalable Anchor Mapping: Leveraging VIO (ARcore) to infer anchor positions based on a few measured control anchors is a clever way to scale up the database construction process without extensive manual surveying for every anchor.
  • Multi-Angle Feature Augmentation is Crucial: The affine invariance enhancement is key to robustness. By generating features from multiple viewpoints, the system can handle significant changes in camera angle, which is a common problem in indoor environments.
  • Spatial Indexing for Speed: The "active anchor" concept and fast spatial indexing dramatically reduce the computational burden of feature matching by narrowing down the search space. This is crucial for real-time applications on mobile devices.
  • Homography for Refined Accuracy: The homography matrix acts as a powerful post-processing step to filter out remaining incorrect matches, boosting the overall localization precision.

Key Results

  • Affine Invariance: FILNet significantly improves feature matching recall under angular deviation, increasing it from 26% to 57% for angles less than 60 degrees.
  • Feature Matching Speed: The fast spatial indexing algorithm reduces average feature matching time from 30.3 ms (K-D tree) to 12.7 ms.
  • Localization Accuracy: The percentage of images with a localization error less than 0.1m increases from 31.61% (baseline) to 55.89% with FILNet.
  • Successful Localization Rate: FILNet outputs localized images for approximately 78% of the test images, compared to 59% for the K-D tree baseline.
  • Robustness to Distance: The method shows good performance across different distances from anchors, with 95% of positioning errors within 0.3m.

Limitations and Future Work

Limitations: * Lighting Sensitivity: The system's performance is significantly impacted by lighting conditions. Low, uneven, or changing light can reduce feature detection and quality, leading to mismatches. * Motion Blur: Camera motion blur from shaking or fast movement can distort features and degrade localization accuracy. * Distance-Induced Distortion: As the distance between the query image and the database increases, perspective differences become more pronounced, potentially causing feature distortion and mismatches.

Future Work: * Improving Lighting Robustness: Investigating methods to ensure effectiveness in insufficient or varying indoor lighting. * Optimizing Indexing Efficiency: Addressing potential reductions in fingerprint database size after screening and improving index efficiency. * Handling Anchor Deployment Issues: Exploring strategies for localization when anchors are deployed sparsely, moved, or removed, potentially by using scene images instead of anchor images. * Expanding Anchor Types: Incorporating objects with salient features already present in the environment as anchors. * Multi-Modal Sensor Fusion: Combining FILNet with other sensors to perceive more anchors and enhance system feasibility. * Anchor Density Analysis: Investigating how anchor distance and density affect localization accuracy.

Summary updated

Abstract

This paper designs a fast image-based indoor localization method based on an anchor control network (FILNet) to improve localization accuracy and shorten the duration of feature matching. Particularly, two stages are developed for the proposed algorithm. The offline stage is to construct an anchor feature fingerprint database based on the concept of an anchor control network. This introduces detailed surveys to infer anchor features according to the information of control anchors using the visual-inertial odometry (VIO) based on Google ARcore. In addition, an affine invariance enhancement algorithm based on feature multi-angle screening and supplementation is developed to solve the image perspective transformation problem and complete the feature fingerprint database construction. In the online stage, a fast spatial indexing approach is adopted to improve the feature matching speed by searching for active anchors and matching only anchor features around the active anchors. Further, to improve the correct matching rate, a homography matrix filter model is used to verify the correctness of feature matching, and the correct matching points are selected. Extensive experiments in real-world scenarios are performed to evaluate the proposed FILNet. The experimental results show that in terms of affine invariance, compared with the initial local features, FILNet significantly improves the recall of feature matching from 26% to 57% when the angular deviation is less than 60 degrees. In the image feature matching stage, compared with the initial K-D tree algorithm, FILNet significantly improves the efficiency of feature matching, and the average time of the test image dataset is reduced from 30.3 ms to 12.7 ms. In terms of localization accuracy, compared with the benchmark method based on image localization, FILNet significantly improves the localization accuracy, and the percentage of images with a localization error of less than 0.1m increases from 31.61% to 55.89%.
BibTeX
@inproceedings{liu2023filnet,
  author = {Sikang Liu and Zhao Huang and Jiafeng Li and Anna Li and Xingru Huang},
  title = {{FILNet: Fast Image-Based Indoor Localization Using an Anchor Control Network}},
  booktitle = {Sensors},
  year = {2023},
  doi = {10.3390/s23198140},
}
Figure