Device-Free Human State Estimation using UWB Multi-Static Radios

TL;DR

The paper tackles indoor localization and activity recognition without requiring users to carry devices by leveraging device-free UWB CIRs processed through a two-stream CNN. It deploys a low-cost multi-static UWB setup with four modules and uses ground-truth from RGB-D cameras synchronized via ROS 2 to train an end-to-end model that maps CIR statistics to location and activities, achieving $L_2$-based localization and high classification accuracy. Key results include a mean localization error of $0.209\,\mathrm{m}$, occupancy counting accuracy of $99\%$, and HAR accuracy of $97\%$, with rapid adaptation to new users via short fine-tuning times. The approach offers privacy-friendly, device-free, and scalable indoor sensing suitable for smart-home and clinical settings, with explicit preprocessing and model-optimization steps to handle environmental variability.

Abstract

We present a human state estimation framework that allows us to estimate the location, and even the activities, of people in an indoor environment without the requirement that they carry a specific devices with them. To achieve this "device free" localization we use a small number of low-cost Ultra-Wide Band (UWB) sensors distributed across the environment of interest. To achieve high quality estimation from the UWB signals merely reflected of people in the environment, we exploit a deep network that can learn to make inferences. The hardware setup consists of commercial off-the-shelf (COTS) single antenna UWB modules for sensing, paired with Raspberry PI units for computational processing and data transfer. We make use of the channel impulse response (CIR) measurements from the UWB sensors to estimate the human state - comprised of location and activity - in a given area. Additionally, we can also estimate the number of humans that occupy this region of interest. In our approach, first, we pre-process the CIR data which involves meticulous aggregation of measurements and extraction of key statistics. Afterwards, we leverage a convolutional deep neural network to map the CIRs into precise location estimates with sub-30 cm accuracy. Similarly, we achieve accurate human activity recognition and occupancy counting results. We show that we can quickly fine-tune our model for new out-of-distribution users, a process that requires only a few minutes of data and a few epochs of training. Our results show that UWB is a promising solution for adaptable smart-home localization and activity recognition problems.

Figures (8)

• Figure 1: Overview of our device-free human-state estimation technique. UWB CIRs are accumulated and used to infer human localization and activity.
• Figure 2: Diagram of UWB module developed at SAIC-Montreal
• Figure 3: ROS 2-based sensor network pipeline. Circles represent “nodes” which collect data or perform computation tasks. Rectangles represent "topics" which are live information stores for nodes to write to and read from to perform their tasks.
• Figure 4: Sample of the trajectory output from our device-free localization model. The ground truth trajectory is in black while the estimated trajectory (temporally smoothed) is in blue. The green dots show the real-world locations of the UWB sensing modules.
• Figure 5: Sample of the trajectory output from our device-free localization model. The ground truth trajectory is in black while the estimated trajectory (temporally smoothed) is in blue. Time on Z-axis.