Multi-Floor Indoor Pedestrian Dead Reckoning with a Backtracking Particle Filter and Viterbi-Based Floor Number Detection.

We present a smartphone-based indoor localisation system, able to track pedestrians over multiple floors. The system uses Pedestrian Dead Reckoning (PDR), which exploits data from the smartphone's inertial measurement unit to estimate the trajectory. The PDR output is matched to a scaled floor plan and fused with model-based WiFi received signal strength fingerprinting by a Backtracking Particle Filter (BPF). We proposed a new Viterbi-based floor detection algorithm, which fuses data from the smartphone's accelerometer, barometer and WiFi RSS measurements to detect stairs and elevator usage and to estimate the correct floor number. We also proposed a clustering algorithm on top of the BPF to solve multimodality, a known problem with particle filters. The proposed system relies on only a few pre-existing access points, whereas most systems assume or require the presence of a dedicated localisation infrastructure. In most public buildings and offices, access points are often available at smaller densities than used for localisation. Our system was extensively tested in a real office environment with seven 41 m × 27 m floors, each of which had two WiFi access points. Our system was evaluated in real-time and batch mode, since the system was able to correct past states. The clustering algorithm reduced the median position error by 17% in real-time and 13% in batch mode, while the floor detection algorithm achieved a 99.1% and 99.7% floor number accuracy in real-time and batch mode, respectively.

LoRaWAN Geo-Tracking using Map Matching and Compass Sensor Fusion.

In contrast to accurate GPS-based localization, approaches to localize within LoRaWAN networks offer the advantages of being low power and low cost. This targets a very different set of use cases and applications on the market where accuracy is not the main considered metric. The localization is performed by the Time Difference of Arrival (TDoA) method and provides discrete position estimates on a map. An accurate "tracking-on-demand" mode for retrieving lost and stolen assets is important. To enable this mode, we propose deploying an e-compass in the mobile LoRa node, which frequently communicates directional information via the payload of the LoRaWAN uplink messages. Fusing this additional information with raw TDoA estimates in a map matching algorithm enables us to estimate the node location with a much increased accuracy. It is shown that this sensor fusion technique outperforms raw TDoA at the cost of only embedding a low-cost e-compass. For driving, cycling, and walking trajectories, we obtained minimal improvements of 65, 76, and 82% on the median errors which were reduced from 206 to 68 m, 197 to 47 m, and 175 to 31 m, respectively. The energy impact of adding an e-compass is limited: energy consumption increases by only 10% compared to traditional LoRa localization, resulting in a solution that is still 14 times more energy-efficient than a GPS-over-LoRa solution.

An Unsupervised Learning Technique to Optimize Radio Maps for Indoor Localization.

A major burden of signal strength-based fingerprinting for indoor positioning is the generation and maintenance of a radio map, also known as a fingerprint database. Model-based radio maps are generated much faster than measurement-based radio maps but are generally not accurate enough. This work proposes a method to automatically construct and optimize a model-based radio map. The method is based on unsupervised learning, where random walks, for which the ground truth locations are unknown, serve as input for the optimization, along with a floor plan and a location tracking algorithm. No measurement campaign or site survey, which are labor-intensive and time-consuming, or inertial sensor measurements, which are often not available and consume additional power, are needed for this approach. Experiments in a large office building, covering over 1100 m², resulted in median accuracies of up to 2.07 m, or a relative improvement of 28.6% with only 15 min of unlabeled training data.

On the use of on-cow accelerometers for the classification of behaviours in dairy barns.

Analysing behaviours can provide insight into the health and overall well-being of dairy cows. Automatic monitoring systems using e.g., accelerometers are becoming increasingly important to accurately quantify cows' behaviours as the herd size increases. The aim of this study is to automatically classify cows' behaviours by comparing leg- and neck-mounted accelerometers, and to study the effect of the sampling rate and the number of accelerometer axes logged on the classification performances. Lying, standing, and feeding behaviours of 16 different lactating dairy cows were logged for 6h with 3D-accelerometers. The behaviours were simultaneously recorded using visual observation and video recordings as a reference. Different features were extracted from the raw data and machine learning algorithms were used for the classification. The classification models using combined data of the neck- and the leg-mounted accelerometers have classified the three behaviours with high precision (80-99%) and sensitivity (87-99%). For the leg-mounted accelerometer, lying behaviour was classified with high precision (99%) and sensitivity (98%). Feeding was classified more accurately by the neck-mounted versus the leg-mounted accelerometer (precision 92% versus 80%; sensitivity 97% versus 88%). Standing was the most difficult behaviour to classify when only one accelerometer was used. In addition, the classification performances were not highly influenced when only X, X and Z, or Z and Y axes were used for the classification instead of three axes, especially for the neck-mounted accelerometer. Moreover, the accuracy of the models decreased with about 20% when the sampling rate was decreased from 1Hz to 0.05Hz.