Multimodal Indoor Localization Using Crowdsourced Radio Maps

TL;DR

This work tackles indoor localization without relying on building floor plans by leveraging crowdsourced radio maps that couple locations with RSS fingerprints. It introduces an uncertainty-aware WiFi localization model that uses CNN-based MAC embeddings on RSS heatmaps and a transformer encoder to produce location estimates with uncertainty, paired with an Extended Kalman Particle Filter for Bayesian fusion with odometry and radio-map priors. Real-world experiments across three buildings show substantial improvements over baselines, with mean errors around 1.7–1.9 m and fusion gains of up to ~43% in the best case, illustrating that crowdsourced radio maps can effectively substitute floor plans in multimodal IPS. The proposed framework enhances robustness to radio-map inaccuracies and sparse coverage, offering a scalable path toward practical, floor-plan-free indoor localization.

Abstract

Indoor Positioning Systems (IPS) traditionally rely on odometry and building infrastructures like WiFi, often supplemented by building floor plans for increased accuracy. However, the limitation of floor plans in terms of availability and timeliness of updates challenges their wide applicability. In contrast, the proliferation of smartphones and WiFi-enabled robots has made crowdsourced radio maps - databases pairing locations with their corresponding Received Signal Strengths (RSS) - increasingly accessible. These radio maps not only provide WiFi fingerprint-location pairs but encode movement regularities akin to the constraints imposed by floor plans. This work investigates the possibility of leveraging these radio maps as a substitute for floor plans in multimodal IPS. We introduce a new framework to address the challenges of radio map inaccuracies and sparse coverage. Our proposed system integrates an uncertainty-aware neural network model for WiFi localization and a bespoken Bayesian fusion technique for optimal fusion. Extensive evaluations on multiple real-world sites indicate a significant performance enhancement, with results showing ~ 25% improvement over the best baseline

Figures (5)

• Figure 1: Conventional IPS uses accurate radio maps for WiFi localization and floor plans in fusion. To fully utilize the noisy crowdsourced radio map, we propose an uncertainty-aware WiFi localization module and a bespoken Bayesian fusion method that replaces floor plans with crowdsourced radio maps in fusion.
• Figure 2: The proposed end-to-end trainable WiFi localization model. Given a WiFi fingerprint list measured by a WiFi-enabled device (the smartphone in the figure), the MAC address part is input-embedded by applying a CNN on its AP-associated RSS distribution map; The RSS part of the fingerprint is embedded via an MLP, and multiplied with the MAC address embedding (c.f. Sec. \ref{['sub:input_embedding']}). The constituted fingerprint embeddings are then fed into the transformer featuring a multi-head attention block in it (c.f. \ref{['sub:multi_head_attention']}). After obtaining the encoded fingerprint features by the transformer, we append two MLPs to regress the location coordinates and quantify the prediction uncertainty (c.f. Sec. \ref{['sub:uncertainty_estimation']}) for subsequent fusion.
• Figure 3: Overall performance of our multimodal IPS.
• Figure 4: Qualitative results of our method in the library (Building C).
• Figure 5: By using the RSS distribution for input embedding, the resultant MAC embeddings are more discriminative where we find physically closer MAC addresses (denoted in similar colors) are closer in the projected 2D feature space.