Robust and Safety-relevant Real-Time Localization
|Duration:||01.06.2016 - 31.10.2018|
|Project Leader:||Prof. Dr.-Ing Horst Hellbrück|
|Staff:||Manfred Constapel (main), Tim Esemann, Marco Cimdins, Mathias Pelka|
Applications which employ autonomous industrial robots, are a possible safety hazard to humans. Robots operate in such safety zones and it is important to surveillance such areas in order to keep humans and machines safe. However, if a human enters a safety relevant zone, it is beneficial to detect this movement and to take appropriate actions.
In this project we aim to develop a robust and safety-relevant real-time localization system. For this, we investigate new kinds of device free passive localization systems to detect reliable human movement in safety-relevant areas.
Student Work in this Research Project
There is always a need for students who wants to participate with our projects. Currently we are offering the following projects. If you have suggestions for an own project, just write an email.
This projects are currently supervised by RosiE:
- Commissioning of Telepresence Robot
- Channel State Information based Localization
- RFID Data Aquisition
- Simulation of radio frequency propagation
- Automatic approach to determine positions of reference points
- Single Anchor Localization (Holger Schwegmann)
- Investigation of different algorithmus for distance estimation (Daniel Amann)
- Underwater Localization (Christian Funda)
- Passive Localization systems (Marco Cimdins)
- Development of a Bluetooth localization system (Thore Kirchmann)
Primilinary results of device-based positioning
In order to make localization systems more robust, we investigate different approaches for localization, including Bluetooth-based fingerprinting or (multi)-lateration-based on distance estimation as well as hyperbolic lateration. Generally, the performance of the localization system increases, the more measurements are available. Such measurements are obtained with fixed reference points which location is known for advances. For each measurement, a reference point is required. Reference points increase the complexity of the localization system.
Our approach in this research project is to decrease the number of the required reference points, but maintain the accuracy and precision of the system. This is possible by integrating multiple antennas into a single reference point. Those antennas serve as virtual reference points. This approach decreases the complexity and costs of a localization system but maintains the accuracy and precision.
This puts high demands on the remaining hardware because the distance measurements towards each antenna should be as precise as possible. If the measurement error of the distance estimation is in the magnitude of the length of antenna geometry, reliable position estimation is not feasible. A remedy for this problem is to employ advanced filtering techniques which correct and refines impure distance measurements. Using advanced signal processing, we can address this issue and still provide reliable position estimation, thus decreasing the complexity and costs of a localization system but still maintain the accuracy and precision.
Evaluation of time-based distance estimation techniques
In an additional investigation, we evaluate different techniques for Two-Way Ranging. Each technique has other properties, including the number of messages, clock drift, and runtime. Additionally, we investigate the difference between free-space distance estimation and distance estimation in closed buildings, like they occur in indoor environments.
The following figure shows the distance estimation error versus the true distance for several ranging techniques.
The systematic error present in the measurements is attributed towards clock drift and other effects. Using advanced filtering techniques to suppress the systematic error, we are able to correct the distance estimation to provide reliable and robust position estimation.
Primilinary results of device-free positioning
The goal of device-free localization systems is the detection and localization of persons that do not carry any devices. When a person is moving within a target area, the signal strength is changed due to changes in the radio frequency propagation. During a training phase, where the target area is vacant from persons, the signal strength is measured, and a threshold is determined.
When this threshold is exceeded during an online phase, a person is detected. The person can be localized by the use of non-linear filters.
The following figure shows a corridor which is supervised by a device-free localization system consisting of 8 sensors (red dots). The blue dot indicates the ground truth position of the person and the black cross the estimated position of the system. Gray circles, so-called particles, are intermediate calculations/samples for the localization algorithm. Ideally, the particles follow the movement of the person and enable precise positioning.
Another important step for the development of a device-free localization system is the understanding of the radio frequency propagation while a person stands at a certain position. We model the impact of the persons via knife-edge diffraction. The following figure shows a person moving through the middle of a radio frequency transmission.
The blue curve are measurements, the orange curve is the result of the simulation.
|||A new localization algorithm based on neural networks , In Proceedings of the 3rd KuVS/GI Expert Talk on Localization, 2018. [bib] [pdf] [abstract]|
Indoor localization plays a major role in a wide range of applications. To determine the location of a tag, localization algorithm is required. In the past, machine learning algorithms were difficult to implement in consumer hardware, but with the advent of tensor processing units, even smartphones are capable to use artificial intelligence to solve complex problems. In this paper, we investigate a machine learning algorithm based on neural networks and compare the result to a linear least squares estimator. We design and evaluate different neural networks. Based on our observation, the neural network delivers poor performance compared to the linear least squares estimator.
|||UWB-based Single Reference Point Positioning System , In ITG-Fachbericht-Mobilkommunikation VDE VERLAG GmbH, 2017. [bib] [abstract]|
Indoor positioning enables new applications, for instance monitoring of goods in smart factories. For such applications, indoor positioning requires cost-effective solutions with high accuracy. State-of-the-art positioning systems are expensive due to high infrastructure and maintenance costs. In this paper we suggest an accurate UWB-based single reference point positioning system using multiple antennas. We compare lateration and hyperbolic lateration as positioning methods and present efficient algorithms for UWB-based single reference point positioning systems. We present theoretical limits based on the Cramer-Rao lower bound and derive an error estimation as well as evaluation results. Our measurements indicate that decimeter accuracy is possible.
|||Minimizing Indoor Localization Errors for Non-Line-of-Sight Propagation , In International Conference on Localization and GNSS, 2018. [bib] [abstract]|
Indoor Localization becomes more important, as it provides additional context for many applications for example in the Internet of Things (IoT). Time-of-flight measurements, as a basis for distance estimation, are susceptible for non-line-of-sight (NLOS) propagation, resulting in large distance errors. Standard least squares solutions to estimate the targets location do not account for NLOS propagation which results in large scale errors. We investigate the difference between L1- and L2-minimization and present a new framework based on a modified RANSAC approach. Additionally, we investigate a Support Vector Machine (SVM) to detect NLOS measurements.We present simulation and measurement results and evaluate our approach. We show that our framework delivers better performance in presence of NLOS propagation compared to plain L1- or L2-minimization.
|||RF-Based Safety-Critical Hybrid Localization , In The Ninth International Conference on Indoor Positioning and Indoor Navigation (IPIN), 2018. [bib] [abstract]|
In safety-critical environments, e.g. paint or machine shops, precise knowledge of positions of persons is important. If an emergency is detected, e.g. a fire or an intruder in the safety-critical area of heavy machinery, emergency shutdown procedures are activated. This requires fine-grained localization with a tag-based localization system, where the person carries a tag. Without a person carrying a tag, precise localization and detection is difficult. In this paper, we propose a RF-based hybrid localization system for safety-critical localization that consists of two major components: a tag-based and device-free subsystem. The device-free subsystem provides coarse-grained localization and monitors a gate area, serving as an entrance towards the safety-critical area where fine-grained localization is required. We propose an architecture of the system, discuss our setup and evaluate typical use cases. Our preliminary evaluation demonstrates that our system detects the correct state of the hybrid localization system with an accuracy of 90%.
|||Sundew: Design and Evaluation of a Model-based Device-free Localization System , In The Ninth International Conference on Indoor Positioning and Indoor Navigation (IPIN), 2018. [bib] [abstract]|
The state-of-the-art in device-free localization systems based on RF-measurements is fingerprinting. Fingerprinting requires reference measurements called fingerprints that are recorded during a training phase. Especially in device-free localization systems, recording of reference measurements for fingerprinting is a tedious, costly, and error-prone task. In this paper, we propose Sundew, a model-based device-free localization system that does not need fingerprinting in the sense of reference measurements but is able to calculate signal strength values at any position and compare it to actual measurements after a simple calibration phase. Sundew — as any device-free localization system — requires a metric for comparison of feature vectors. In this paper, we investigate the influence of nine different distance metrics on the positioning accuracy. Simulations and measurements show that our suggested model-based device-free localization system works best with the L 1 distance metric. Sundew estimates 90% of positions in a 2.5m x 2.5m grid correctly, independent of the orientation of the person in the target area.
|||Introduction, Discussion and Evaluation of Recursive Bayesian Filters for Linear and Nonlinear Filtering Problems in Indoor Localization , In The Seventh International Conference on Indoor Positioning and Indoor Navigation, 2016. [bib] [abstract]|
Linear and nonlinear filtering for state estimation (e.g. position estimation or sensor fusion) for indoor positioning and navigation applications is a challenging task. Sensor fusion becomes more important with cost-effective sensors being readily available. However, state estimation with recursive Bayesian filters for sensor fusion and filtering are difficult to apply. We present an overview for the general Bayesian filter and derive the most commonly used recursive Bayesian filters, namely the Kalman, extended Kalman and the unscented Kalman filter along with the particle filter. The later Kalman filters are extension of the original Kalman filter, which are able to solve nonlinear filtering problems. The particle filter is also able to solve nonlinear filtering problems. We evaluate the recursive Bayesian filters for linear and nonlinear filtering problems for sensor fusion from relative dead reckoning positioning data and absolute positioning data from an UWB positioning system. We discuss and evaluate performance and computational complexity and provide recommendations for the use case of the recursive Bayesian filters.
|||Investigation of Anomaly-based Passive Localization with Received Signal Strength for IEEE 802.15.4 , In The Seventh International Conference on Indoor Positioning and Indoor Navigation, 2016. [bib] [abstract]|
Localization has important applications, for instance intrusion detection and elderly care. Such applications benefit from Device-free passive (DfP) localization systems, which employ received signal strength measurements (RSSM) to detect and track entities that neither participate actively in the localization process nor emit signals actively. RSSMs include received signal strength indicator (RSSI), energy detection (ED) and link quality indicator (LQI) measurements. This paper compares different packet-based RSSMs for DfP localization and presents detection results of a DfP anomaly-based detection system employed by IEEE 802.15.4 compliant devices. Furthermore, we investigate techniques for anomaly detection with continuous RSSI measurements.
|||Evaluation of time-based ranging methods: Does the choice matter? , In 14th Workshop on Positioning, Navigation and Communication, 2017. [bib] [abstract]|
Positioning is useful in a number of applications, for instance smart home, smart factory and health care applications. Time-based ranging methods for positioning are the state-of-the-art but require precise timestamping. Sophisticated ranging methods compensate sources of errors, for instance clock drift caused by a crystal or an asymmetrical measuring principle, to provide precise timestamping. So far, no comprehensive study of different time-based ranging methods using the same hardware and the same evaluation setup was carried out. Consequently, we discuss, implement and evaluate five time-based ranging methods, including Two-Way Ranging, Double Two-Way Ranging, Asymmetrical Double-Sided Two-Way Ranging, Symmetrical Double-Sided Two-Way Ranging and Burst Mode Symmetric Double-Sided Two-Way Ranging. We evaluate accuracy, precision, robustness and run time for the ranging methods and answer the question if the choice of the time-based ranging method matters.
|||Anomaly-based Device-free Localization with Particle Filtering , In Workshop on Dependable Wireless Communications and Localization for the IoT, 2017. [bib] [abstract]|
In the Internet of Things (IoT), devices, e.g. sensors or actuators, transmit packets to transfer data. For the IoT localization information is crucial, as it provides additional context for the data. We envision that devices in the IoT know their position and on receipt of a packet, the received signal strength is measured. This measurement is used to build a device-free localization (DFL) system to improve the dependability of the IoT system. DFL systems are able to detect and track persons within a target area that neither wear a device nor participate actively in the process of localization. This work presents an anomaly-based DFL system that measures if a person affects the radio frequency (RF) propagation and determines the position with a particle filter. In our 65m 2 indoor testbed, we employ eight IEEE 802.15.4 compliant wireless transceivers and estimate the position of a person with a median localization error of 1.4m.
|||Modeling Received Signal Strength and Multipath Propagation Effects of Moving Persons , In 14th Workshop on Positioning, Navigation and Communication, 2017. [bib] [abstract]|
Device-free localization (DFL) systems detect and track persons without devices that participate in the localization process. A person moving within a target area affects the electromagnetic field that is measured by received signal strength (RSS) values. Consequently for DFL systems modeling of RSS is important and still an open issue. In this paper, we develop a simple model for prediction of RSS values in a setup with transmitter and receiver devices, a person and multipath propagation. We design and implement the model as a superposition of both, knife-edge diffraction to account for the change made by the person, and, propagation effects such as multipath propagation that result in reflection and path loss including the antenna characteristics. We evaluate our model in comparison with real measurements in various setups with and without multipath propagation. We achieve an accuracy that is close to our hardware limitations, which is the resolution of the measured RSS values of the receiver.
|||Impact of the antenna orientation for distance estimation , Technical report, Technische Universität Braunschweig, 2018. [bib] [pdf] [abstract]|
Indoor localization is important for a wide range of use cases including industrial, medical and scientific applications. The location accuracy is affected by the localization algorithm and the quality of the measurements as input for the algorithm. Many indoor localization systems employ ultra-wideband distance measurements, as they offer high accuracy and are cost effective. One of the methods for distance measurement is twoway ranging. This paper investigates the impact of the antenna orientation on the distance measurement based on symmetrical double-sided two-way ranging. We show that up to 0.25m of the measurement error is attributed to the orientation of the antennas. We provide explanations and suggest solutions to reduce the effect.
|||Comparison of Antenna Types and Frequency Bands for Radio-based Device-free Localization , Technical report, Technische Universität Braunschweig, 2018. [bib] [pdf] [abstract]|
Radio-based device-free localization systems measure effects on radio signals e.g. signal strength variations to locate objects or persons in a target area. Such systems detect and track persons that do not participate in the localization process. Models for calculating the radio signal propagation are key for the performance in device-free localization systems. Received signal strength (RSS) is simple to measure. However, it is susceptible to changes in the environment and multipath propagation. In this paper, we compare PCB antennas to a circularly polarized cloverleaf antenna and measurements in the 2.4 GHz with measurements to the 868MHz ISM band. We investigate especially if a circularly polarized cloverleaf antenna is resilient against multipath propagation. Our preliminary results demonstrate that our model is suitable to the 868MHz band and the use of the 868MHz band increases the area where a person affects the RSS. The use of a circularly polarized cloverleaf antenna does not help to avoid multipath propagation.
|||Evaluation of Bluetooth Positioning for Medical Device Tracking , GRIN (T. M. Buzug et. al., ed.), 2017. [bib]|
|||Investigation of Anomaly-based Passive Localization with IEEE 802.15.4 , Technical report, RWTH Aachen University, 2016. [bib] [pdf]|