key: cord-0058622-32dd3fxa authors: Sonnessa, Alberico; Saponaro, Mirko; Alfio, Vincenzo Saverio; Capolupo, Alessandra; Turso, Adriano; Tarantino, Eufemia title: Indoor Positioning Methods – A Short Review and First Tests Using a Robotic Platform for Tunnel Monitoring date: 2020-08-19 journal: Computational Science and Its Applications - ICCSA 2020 DOI: 10.1007/978-3-030-58811-3_48 sha: c02cf247800497c9f22e9024aad54d9ba2441a49 doc_id: 58622 cord_uid: 32dd3fxa The aim of this work is to provide a review of the main indoor positioning methodologies, in order to evidence their strengths and weaknesses, and explore the potential of the integration in an Unmanned Ground Vehicle built for tunnel monitoring purposes. A robotic platform, named Bulldog, has been designed and assembled by Sipal S.p.a., with the support of the research group Applied Geomatic laboratory (AGlab) of the Politecnico di Bari, in the definition of the data processing pipeline. Preliminary results show that the integration of indoor positioning techniques in the Bulldog platform represents an important advance for accurate monitoring and analysis of a tunnel during the construction stage, allowing a fast and reliable survey of the indoor environment and requiring, at this prototypal stage of development, only a remote supervision by the operator. Expected improvements will allow to carry out tunnel monitoring activities in a fully autonomous mode, bringing benefit for the safety of people involved in the construction works and the accuracy of the acquired dataset. The aim of the research presented hereafter is to provide a short review of the main indoor positioning methods, by showing the preliminary results of an industrial application in the field of civil engineering, developed by Sipal S.p.a. and coordinated by the research group AGlab at the Politecnico di Bari, aimed at exploring the opportunity to deploy Unmanned Ground Vehicles for tunnel monitoring purposes. To achieve this, a remote-controlled prototype robotic platform, equipped with sensors, has been designed and assembled. Field tests were carried out to assess the performance of the system and are still ongoing. The first results indicate that the integration of indoor positioning techniques in the inspection and monitoring activities can be a valid support in the tunnel construction stages. The accurate mapping of the environment in which we live, along with the precise positioning of people and objects in it, still represents one of the most challenging issues for the scientific community. Nowadays, high-precision mapping techniques, that encompass satellite and ground-based sensors, allow to get very detailed 3D models of the territory and the built environment. Current positioning and navigation techniques, which mainly rely on geodetic GNSS (Global Navigation Satellite System) receivers, permit to estimate positions of a sensor with near-millimeter accuracy. As for the consumer market, mobile phones made real-time positioning within everyone's reach. Thanks to increasingly more advanced sensors and algorithms embedded in our devices and the availability of web mapping/navigation services (e.g. Google Maps, Apple Maps), we can easily locate ourselves and navigate towards every destination. Outdoor real-time precise positioning needs a good visibility of the satellite constellations (open sky conditions) and a web access. At indoors, the level of degradation of the satellite signals makes the civiliangraded accuracy of GNSS inadequate for many purposes [1] . However, the estimation of a sensor indoor position and/or motion finds application in different contexts, such as augmented/virtual reality, autonomous robot control [2] and civil engineering. The improvements in the reliability of outdoor positioning methodologies, do not correspond to same advances in indoor positioning systems (IPS). Simultaneous Localization And Mapping (SLAM) problem for indoor locations, that is the simultaneous estimation of the state of a platform equipped with on-board sensors and the building of a model (map) describing what is observed by the sensors [3] , is still an open issue. Consumer market solutions have been recently implemented by Apple in its highend iPad tablet, that couple cameras and laser sensors. Interesting attempts aimed at freely making available indoor maps is the use of an Open Source platform to support indoor navigation, based on the OpenStreetMap tool, as showed in [4] , and the opportunity provided by Google to its partners of integrating indoor maps in Google Maps. • Mining activities. • Operation in dangerous environments. With regards to the case study, in Table 1 are summarized the main positioning requirements in underground construction. In accordance with the conditions indicated in the table it is therefore necessary to design an indoor system that guarantees the performance of the positioning. The position information of an Unmanned Vehicle can be provided by using various methodologies exploiting different sensors and technologies (or a combination of them). Main IP techniques will be shortly described in the following section along with main advantages and limitations. IPS can be classified starting from the principles on which they are based, namely electromagnetic waves (visible, infrared, microwave and radio spectrum), mechanical waves (audible/ultra-sound) and inertial navigation (accelerometers and gyroscopes) [6] . Despite the difficulties of performing IP -especially where scenarios rapidly change for the presence of "moving things" (e.g. workers, construction machinery)depending on possible Non-Line-of-Sight (NLoS) conditions, multipath from walls, ceilings and objects, signal attenuation/scattering -nevertheless, indoor locations can simplify navigation and positioning tasks, because of: • Slow dynamics (with reference to the outdoor environment). • Limited areas to be covered. • Availability of facilities (electric power, web access, walls for target installation) • Very limited weather effects. Main IP techniques are listed in Table 2 . Optical IP uses cameras as main sensor. With the improvement of their performances in terms of resolution of collected images/data and speed in their acquisition, the accuracy of this method has significantly improved in the last years. The retrieving of 3D information about the investigated environment is based on a photogrammetric approach. As discussed in [7] , the scale of the photogrammetric model can be obtained integrating a Time Of Flight (ToF) range camera. Different references can be used to carry out the sensor positioning, according to [6] , that is: Targets/patterns when mounting of reference markers is not indicated or possible. 3. Reference from Deployed Coded Targets aimed at a) simplifying the automatic identification of corresponding points assigning a unique code to each marker, b) providing the system scale. 4. Systems without Reference that relies on the direct tracking of objects. Illuminated targets are often used to improve algorithmic robustness. Cameras can be used to simultaneously map the surrounding environment and estimate motion, i.e. performing a Visual SLAM. According to [8] , this solution can rely on geometry by getting geometric constraints from images for motion estimation, or, following the recent trends, using deep learning to autonomously reconstruct and navigate in an unknown location. An innovative approach to implement Visual SLAM, based on the integration of standard and event cameras with IMU (Inertial Measurement Unit), has been presented in [2] . This method overcomes typical limitations affecting standard cameras, (motion blur and low dynamic range) integrating event cameras, where pixels operate autonomously, transmitting their changes of intensity/time they occur/pixel space-time coordinates. These sensors produce little amount of information when motion is limited. The authors in [2] introduced a pipeline that combines events, standard frames, and inertial measurements to compute reliable and accurate state estimation, applied for onboard state estimation of an autonomous quadrotor. This approach led to an accuracy enhancement of 130% if compared to the use of the event camera only, and 85% over standard-frames-only visual-inertial systems, although still being computationally manageable. Main advantages and limitations of this technique are summarized in Table 3 . Infrared (IR) light is characterized by a wavelength longer than those of visible light, thus making this technology less invasive with respect to the IP techniques based on visible light. According to [6] , the main methods of using infrared signals are: a) artificial light sources b) infrared imaging based on thermal radiation c) use of active beacons. Need for providing a constantly updated database (1) Must be generally coupled with other sensor for SLAM purposes Cost 1. Artificial Infrared Light sources using active infrared sources and infrared sensitive cameras. An application in the gaming field is the Kinect (V1) embedded in the Xbox console, where the 3D shape is retrieved from the distortion of a pseudo random pattern of structured IR light dots captured with the infrared camera. Though developed in the video gaming world, Kinect has been wide spreadly used in the SLAM field thanks to the KinectFusion and Kintinuos algorithms [9] [10] [11] . 2. Imaging using Natural Infrared Radiation, namely passive IR localization methods. Sensors operational in the thermography region are capable to acquire a passive image of the environment using natural thermal emissions. 3. Active Beacons based on infrared receivers located at known positions and mobile beacons whose positions are not known. A sub-meter precision can be reached when several receivers are employed in each room. Main advantages and limitations of this technique are listed in Table 4 . The sound wave is a particular wave in which the perturbation is the pressure variation induced by a vibrating body in the surrounding medium (usually air or building material). The use of sound waves for IP purposes is based on locating mobile nodes through multilateration and TOA (Time of Arrival)/TDOA (Time Difference of Arrival) that is the multiple distance measurements to static nodes mounted permanently in a given environment. Most widely known techniques are [12] : -Active Bat A network of receivers with a centralized control unit are used to process the transmitters signal (ultrasound) and estimate their position. A number of active devices (beacons, emitting ultrasound)), called 'Cricket nodes' [13] , establish the positioning network and are used by the target (listener) for position determination. Mixed ultrasound-RF system, where a number of reference nodes emit in series RF and ultrasound signal, used by other nodes to compute distance to reference node. Table 5 . Due to its wide deployment in many indoor environments, WLAN (Wireless Local Area Networks), known as Wi-Fi, is employed to estimate the position of a mobile device in a range up to about 100 m. To date, empirical models for WLAN-based indoor positioning are preferred [15] , such as fingerprinting (FP) [16, 17] hereafter described. As first step, signals are collected offline in order to build a database of discrete grid of locations, or reference point (RP), namely the radio-map. A signature is generated from the collected RF signals. The next stage, that is the online phase, make use of the radio-map for navigation via classification algorithms, or novel methodologies based on machine learning [18] , and deep learning [19] [20] [21] [22] . The accuracy of FP is strictly related to the number of calibration points, ranging from one to tens of meters. Main FP-based IPS are: -FP-Based IPS Based on RSS This method employs received signal strength (RSS) measurements as fingerprints for radio-map creation and online navigation. This approach is based on the channel estimate, which defines an indoor location using multipath attenuation and further physical phenomena such as refractions and reflections, characterizing this environment with detailed information [18, 19, 22] . The channel state information (CSI) can be used as FP. Table 6 . Expensive for large environments Accuracy (few centimeters) [14] Variable accuracy as sound speed depends on temperature and humidity 4.5 Radio Frequency Identification RFID (Radio Frequency IDentification) IP methods relies on a reader, capable of obtaining through RF the data from active, passive or semi-passive electronic tags storing an ID, used to locate the RFID tag. A standard application able to manage a large number of tags is in the product tracking aimed at avoiding stealing in stores, with readers at known locations and tags on the product to be monitored. The accuracy and the range in RFID may vary significantly, depending on whether they are active or passive and the density of devices, with some researches achieving around-centimeters accuracies [23] . The typical case of RFID positioning using ToA, is based on measurements from a single tag; positioning is obtained combining the Angle of Arrival (AoA) measurements from that tag [1] . Main advantages and limitations of this technique are shown in Table 7 . A review of the state of the art in Ultra-Wide Band (UWB) positioning has been recently performed in [24] . UWB is defined as a RF signal characterized by a bandwidth greater than 500 MHz, usually operated in the 3.1-10.6 GHz range. According to [25] , UWB positioning methods can be grouped into two main classes, i.e. FP-based and geometric methods. Geometric methods are hereafter described, because comprehensive FP database are difficult to build. Signal power loss due to reflection, absorption, refraction, scattering, interference, and multipath Privacy concern Unique ID can be used for security purposes Data management and configuration when a large number of devices is deployed -TOA and TDOA-Based Methods The TOA method is based on computing of the range transmitter -receiver by estimating of the Time Of Arrival of signals, thus requiring the precise synchronizations between the transmitter and receiver. The TDOA method locates the receiver using the Time Difference Of Arrival computed between the receiver and multiple transmitters, not necessitating their synchronization [26] . The AOA-based approaches first make use of the angle of the signal received from the transmitter, and then estimate the receiver location by using the intersection of the angle line from the transmitters [25] . Such approaches don't require clock synchronization and use a small number of anchors nodes, nevertheless necessitating the coupling of nodes with an antenna array. Geometric-RSSI (Received Signal Strength Indicator) methods relies on the measurement of the strength of the signal acquired by the receiver and transform it into a range observation between the transmitter and receiver. The performance of RSS methods is usually low if compared to TOA and TDOA approaches [27] , but they are relatively low cost, and can deploy a large number of nodes. Main advantages and limitations of this technique are summarized in Table 8 . GNSS receivers hardly track signals in challenging settings such as urban canyons or at indoors, where signal fades [28] . This has boosted the research in the field of so-called highly sensitive and assisted GNSS (HSGNSS-AGNSS) receivers. AGNSS receivers use an additional data link (e.g. internet connection) for providing information about satellite ephemeris, almanac, differential corrections and timing information, usually directly acquired from the GNSS constellations. High-Sensitive GNSS (HSGNSS) receivers deploy further methods, not deepened here, aimed at increasing the signal reception in indoor environments. However, performance of indoor GNSS technologies are not yet comparable to the level of accuracy reachable in outdoor settings. Even if indoor GNSS do not provide the needed accuracy for pedestrian navigation in a closed environment, this sensor can be integrated in an IPS with an inertial platform, in order to supply sparse coordinate updates. Table 9 . The pseudolite system represents a powerful indoor positioning system, due to the capability of covering large areas and provide accurate positioning solution [29] . Pseudolites (pseudo-satellites) are basically ground-based beacons capable to generate pseudo-codes analogous to GNSS codes. This solution also embeds mobile rovers whose positions are estimated by measuring the distance between rovers and pseudolite beacons, generally positioned at known locations [6] . Pseudolites can cover wide areas (tens of kilometers); the main restriction in the yr usability is related to NLoS conditions between pseudolites and rovers. The use of a combined pseudolite receiver enables the acquiring and tracking of both GNSS and pseudolite signals, thus permitting a seamless transition from outdoor to indoor settings. Main advantages and limitations of this technique are listed in Table 10 . Dead Reckoning (DR) relies on the estimation of the current location of a target, computed from an already known position (fix), and taking advantages on the measurements of physical quantities used for defining its movement, (e.g. path and speed) [1] . The DR technique based on Inertial Navigation Systems (INS) encompass the integration of an Inertial Measurement Unit (IMU) with a computational platform. No easy solution for ambiguity resolution An INS estimates the position, velocity and orientation quantities from the IMU, which embeds gyroscopes, accelerometers, and/or a magnetometer aimed at measuring the direction and strength of the magnetic field. Due to significant drift problems, INS is commonly integrated with additional sensors which provide information on the system position. Main advantages and limitations of this technique are evidenced in Table 11 . Polar systems measure distance and angles, retrieving the position of a target object using the three-dimension vector between sensor and objects. Most widely known polar systems are Total Stations (TS) and Laser Scanners (LS). Basically, TS and LS rely on the same measuring principles, where angles are measured by means of digital bar-codes embedded in the sensor, and an Electronic Distance Meter (EDM) measures the distance from the instrument to a target (typically a retroreflective prism) or a surface by timing the round-trip of a pulse of light. The signal reflected by the surface is detected by a sensor connected with the emitter. The instrument automatically scans the surfaces in its field of view and is capable to acquire millions of point per second [30] . These instruments mainly differ in terms of accuracy (TS can reach an accuracy of few millimeters) speed of acquisition and number of acquired point per seconds (LS can acquire up to one million of point/sec). IP Polar systems-based tools, relying on high-precision rugged sensors/instruments developed for working in severe conditions, are particularly suited for civil engineering applications. Main advantages and limitations of this technique are showed in Table 12 . The project is co-funded by the European Union-European Regional Development Fund POR Puglia 2014/2020 and the Puglia Region. A section of the project involves the development of hardware and software tools to be installed on board aerial and ground UVs, aimed at integrating or replacing traditional survey techniques used for collecting and analyzing key data in the different stages of a construction site. Within this activity, it is being implementing a platform named Bulldog to be used for tunnel monitoring purposes [31] , that is: -The estimation of the over/under-excavated sections of the tunnel with respect to the type section. -The survey of the excavation head, during the tunneling works, for controlling its stability and identifying the most suitable consolidation systems. -The automatic identification of under-excavated areas that requires further work to reach the minimum theoretical tunnel section. The Bulldog platform (Fig. 1 ) exploits IP Polar system-based techniques. The hardware of the platform consists of: -A remotely controllable Unmanned Ground Vehicle (UGV), (max speed: 30 km/h, max payload: 130 kg) equipped with two Optic cameras and two IR cameras. -A Trimble SX10 scanning station, that integrates the features of a TS (angular accuracy: 1'', distance accuracy: 1 mm + 1.5 ppm) and LS (3D point position accuracy @100 m: 2.5 mm). -A GEO-Laser AD-12 automatic tripod, for automatic levelling of the surveying instrument. At this stage of its development, Bulldog must be remotely driven towards the target area inside the tunnel. As first step, the platform carries out high-precision localization operations computing its position by mean of the TS. The sensor position is obtained by measuring some retro-reflective prisms of known location installed in the tunnel. After the preliminary IP operations, Bulldog is able to perform a scan of the target area in LS mode, providing a point cloud with an average centimeter spacing. The acquired data are processed and analyzed using specific tools developed on purpose in Python language, able to extract cross-sections of the tunnel and related construction characteristics and to estimate the over/under-excavated sections of the tunnel with respect to the typical section (Fig. 2) . Bulldog has been designed considering the requirements in terms of needed accuracy for the UGV positioning and reconstruction of the 3D model (some millimeters) of the surveyed area, and the possibility of easily upgrading the system by replacing or adding new SW features and/or sensors. Next steps expect the implementation of SLAM algorithms in the system, in order to make Bulldog capable of autonomously detect obstacles and move inside the construction area from its initial position, simply providing to the system the coordinates of an end point. To this aim, a Velodyne VLP-16 LS and an IMU will be integrated in the platform. While not as accurate as SX10, Velodyne LS can be used for navigation, being able to quickly acquire a reliable (about 3 cm of accuracy) point cloud, useful for a fast and continuous mapping of the indoor environment. Therefore, the lack of efficient realtime 3D model generation methods of as-built 3D indoor environment is still an open issue. An IMU will be also integrated to better reconstruct the Bulldog trajectories. In this prototyping stage, SLAM algorithms are verified using ROS (Robot Operating System), a set of open source software libraries and tools for building robot applications [32] , installed on TurtleBot, a low-cost robot kit with open-source software, equipped with a 2D or 3D distance sensor, a simplified IMU and microcomputer [33] . ROS+TurtleBot enable the testing and improvement of new features, allowing the speeding up of the system development. Positioning and mapping activities in an indoor environment represent, in some cases, very difficult tasks. Nevertheless, high-accuracy IPSs have become in high demands in recent years, due to the increasing need for precise indoor localization in many sectors, ranging from the assistance for aged people to pedestrian navigation in large public buildings, to autonomous driving. In the field of civil engineering, accurate IP and SLAM are required especially in potentially dangerous environments, as construction sites, where traditional measurement techniques could be replaced by fully automated procedures, avoiding possible injuries to operators. In this work, different IP methodologies have been described in order to provide an overview of the main advantages and limitation of each technique, and an application of IP polar system-based method to an integrated robotic platform for monitoring tunnels has been presented. As discussed above, the integration of IP techniques in the Bulldog platform represents an important advance for accurate monitoring and analysis of a tunnel during the construction works, allowing a fast and reliable survey of indoor environments and requiring, at this stage of development, only a remote supervision by the operator. Expected improvements will allow to carry out monitoring activities without the need of a surveyor, bringing benefit for the safety of the construction area and accuracy of the acquired dataset by reducing possible errors due to the human factor. A Meta-Review of Indoor Positioning Systems Ultimate SLAM? combining events, images, and IMU for robust visual SLAM in HDR and high-speed scenarios Past, present, and future of simultaneous localization and mapping: toward the robust-perception age An open source platform for indoor navigation: application to the Faculty of Requirements for positioning and navigation in underground constructions Indoor Positioning Technologies Kinect V2 and RGB stereo-cameras integration for depth map enhancement Deep learning for visual SLAM in transportation robotics: a review KinectFusion: real-time dense surface mapping and tracking KinectFusion: real-time 3D reconstruction and interaction using a moving depth camera Kintinuous: Spatially extended kinectfusion A state-of-the-art survey of indoor positioning and navigation systems and technologies The cricket indoor location system Recent advances in indoor localization: a survey on theoretical approaches and applications Indoor positioning via WLAN channel state information and machine learning classification approaches RADAR: an in-building RF-based user location and tracking system He Horus WLAN location determination system Fast prototyping of an SDR WLAN 802.11b Receiver for an indoor positioning systems CSI-based fingerprinting for indoor localization: a deep learning approach CSI phase fingerprinting for indoor localization with a deep learning approach Deep learning-based indoor localization using received signal strength and channel state information ConFi: convolutional neural networks based indoor Wi-Fi localization using channel state information Evolution of indoor positioning technologies: a survey Experimental evaluation of a UWB-based cooperative positioning system for pedestrians in GNSS-denied environment Precise indoor positioning using UWB: a review of methods, algorithms and implementations High-accuracy TDOA-based localization without time synchronization UWB and MEMS-based indoor navigation Evaluation of high sensitivity GPS receivers A new indoor positioning system architecture using GPS signals High-resolution geomatic and geophysical techniques integrated with chemical analyses for the characterization of a Roman wall Piattaforma HW/SW per la gestione dei Cantieri Tecnologici per Infrastrutture Civili Acknowledgements. This research is funded by the project "Technological Construction Site for Military and Civil Infrastructures/Cantiere Tecnologico per Infrastrutture Militari e Civili." (Unmanned Vehicles and Virtual Facilities), co-financed by the European Union-European Regional Development Fund POR Puglia 2014/2020 and Puglia Region.