Smartphone-Based Portable Pedestrian Indoor Navigation, Tesine di Maturità di Ingegneria Aerospaziale

Scarica Smartphone-Based Portable Pedestrian Indoor Navigation e più Tesine di Maturità in PDF di Ingegneria Aerospaziale solo su Docsity! 311 STEPPING – Smartphone-Based Portable Pedestrian Indoor Navigation Christian Lukianto1, Harald Sternberg 2 1, 2 Department of Geomatics, HafenCity University Hamburg Hebebrandstraße 1, 22297 Hamburg, Germany christian.lukianto@hcu-hamburg.de, herald.sternberg@hcu-hamburg.de KEY WORDS: Navigation, GPS/INS, IMU, Multisensor Fusion, Algorithms, Digital Sensor Systems ABSTRACT: Many current smartphones are fitted with GPS receivers, which, in combination with a map application form a pedestrian navigation system for outdoor purposes. However, once an area with insufficient satellite signal coverage is entered, these navigation systems cease to function. For indoor positioning, there are already several solutions available which are usually based on measured distances to reference points. These solutions can achieve resolutions as low as the sub-millimetre range depending on the complexity of the set-up. STEPPING project, developed at HCU Hamburg - Germany aims at designing an indoor navigation system consisting of a small inertial navigation system and a new, robust sensor fusion algorithm running on a current smartphone. As this system is theoretically able to integrate any available positioning method, it is independent of a particular method and can thus be realized on a smartphone without affecting user mobility. Potential applications include -- but are not limited to: Large trade fairs, airports, parking decks and shopping malls, as well as ambient assisted living scenarios. 1. INTRODUCTION Presently, most smartphones are equipped with GPS receivers, which, in combination with a map application turn the phone into a sophisticated pedestrian outdoor navigation system. However, once the user of such a navigation system enters a region of insufficient satellite signal coverage, e.g. an urban canyon or even a building, this navigation system ceases to function immediately. At the same time, research on indoor positioning systems has effected the output of various solutions for the indoor positioning problem: how to determine the position with the desired accuracy in the absence of viable satellite signals. Various techniques achieve different accuracies and ranges, depending on the set-up and the complexity of the system. Solutions are based on light (laser/infrared or optical markers), ultrasound and (high-frequency) electromagnetic waves, such as rfID, WiFi, Bluetooth or UWB systems. A comprehensive list and comparison of achievable ranges and accuracies can be found in (Mautz, 2009). This paper proposes a research project, which recognizes the ever increasing processing power and degree of integrated features of mobile IT devices (i.e. smartphones). It aims at combining these smartphones with available indoor navigation infrastructure to form a robust pedestrian indoor navigation system. The project's goals are presented followed by a detailed discussion of the individual components and concepts and a description statement of the current state of development. Archives of Photogrammetry, Cartography and Remote Sensing, Vol. 22, 2011, pp. 311-323 ISSN 2083-2214 312 Fig. 1 STEPPING System Concept - External navigation information is integrated by the smartphone to support the drifting INS. A geo server supplies floor plan and infrastructure data upon entering the building 2. PROJECT DESCRIPTION Any kind of navigation system requires a continuous, long-term stable position update. Outdoors, this is achieved by using the permanently available satellite signals of global navigation satellite systems (GNSS) such as GPS, GLONASS or GALILEO. Indoor navigation conditions usually imply the absence of, or at least only partial coverage by the aforementioned satellite signals. As, under these conditions, a continuous position update is difficult to guarantee using standard satellite receivers, an inertial navigation system (INS) is used to provide the continuous position update. However, owing to accumulated measurement errors, the position estimates provided by the INS are only valid in the short term. INS require external support information to correct these errors and to ensure the continued viability of the position estimate (Grewal, et al., 2007). STEPPING project acknowledges the fact, that there may not always be the same particular kind of indoor navigation infrastructure available. However, it assumes that, as the user moves through the different (indoor) environments, there is always some kind of navigation infrastructure installed. The information provided by the external infrastructure is then used to support the INS while at the same time making the system independent of a particular navigation infrastructure type. The smartphone, already equipped with a plethora of wireless communication interfaces, is the optimal choice of platform, as it interfaces effortlessly with the available navigation infrastructure (see Figure 1). Christian Lukianto, Harald Sternberg 315 used to stabilize gyroscope drift and the altitude channel respectively. Also, the magnetometer is used during initialization. Owing to the inherent short-term validity of INS position estimates, the INS is supported by the current global optimal position estimate from the master filter as long as the global QoL is better than local INS QoL. 3.2.2.Step Counter This subfilter can be used in two ways. In the first mode, the filter operates as a velocity estimator. Dynamic models, describing the human walk and the relationship between step frequency and velocity have been discussed in the literature (Jahn, et al., 2010). The Nokia N900 smartphone has a built-in accelerometer sensor, which in this mode, provides an INS- independent reading used for step-counting. The resulting velocity estimate can be used to support the INS velocity estimate, or to detect stationary segments. The combination of the velocity estimate with the orientation estimate from the INS constitutes a second mode of operation. The so called pedestrian dead reckoning (PDR) algorithm (Widyawan, et al., 2008) computes position increments from the velocity and direction of motion. From a known point of origin, velocity is integrated to yield a distance estimate. Since the direction of motion is known, a new position estimate can be computed adding the increment to the known point of origin. 3.2.3.GPS/Cell-based Positioning The smartphone includes a GPS receiver. Given sufficient satellite signal coverage, this module provides a long-term valid and accurate position. The module is directly connected to the 3G/2G communications module which in itself can provide cellular network based position information. The combination of both considerably reduces the time to first fix and usually yields a valid GPS position after mere seconds. Yet, even if there is no valid fix available, cell-based positioning can still be used to aid the master filter. The accuracy is dependent on the density of the surrounding cellular network masts. Also, the GPS-related values for dilution of precision (DOP), which are supplied by the antenna with every reading, can be used directly in the computation of QoL. 3.2.4.Particle-Filter Particle filters are also known as sequential monte-carlo methods (SMC). They belong to a group of stochastic methods to estimate the states of dynamic processes which have the following properties: Only their mean dynamic behaviour is known and not all states are observable (Wendel, 2007). A particle filter treats the state vector as a random variable whose probability density function is approximated numerically. At each propagation step, states below a certain probability threshold are eliminated by including (not necessarily continuous) measurements until the filter converges on the most likely solution, i.e. the state with the highest probability. Stepping – smartphone-based portable pedestrian indoor navigation 316 In this particular implementation, all possible positions inside a (known) building are modelled on a floor plan. If then, at each propagation step, some (partial) state observations are added to the filter (e.g. some distance measurements in combination with a direction estimate), more and more improbable positions can be eliminated until the filter converges. This is also possible for a moving target. 3.2.5.Discrete WiFi-Fingerprinting WiFi networks are already used for positioning in many research activities. A major difference between two groups of techniques is, if the absolute position of the WiFi nodes is known or not. In the first case, methods from the field of cell-based positioning are employed, such as triangulation or the cell of origin (COO) method. In the second case, the absolute position of the nodes only plays a minor role: Here, the received signal strengths (RSS) are measured and recorded for every point on a grid covering the floor plan. For positioning, the current measured RSS is compared with the reference values. The process is called WiFi-fingerprinting and is very elaborate and expensive. Also, it is very sensitive towards small changes in the environment (e.g. a displaced piece of furniture) or the equipment used, as RSS values are not unique over a range of node/receiver combinations. Fig. 3 Sample QR-Code encoding textual information For that reason, a more robust method will be employed, as described by (Sayrafian-Pour, et al., 2008). The method still uses RSS but only compares relative strengths. Only the orders of magnitude of the RSS are compared. Using relative instead of absolute signal strengths makes the method independent of the used node/receiver combination and thus more robust. It is to be investigated further, if the use of certain models for signal attenuation by the surrounding walls is able to render the fingerprinting process unnecessary. Christian Lukianto, Harald Sternberg 317 3.2.6.Further Subfilters As mentioned earlier, the number of employed subfilters is only limited by the smartphone's system resources. However, as not all subfilters run at the same time and have different requirements regarding power, memory or CPU cycles, there are several other subfilters possible: 3.2.6.1.Door-identification using magnetometer The magnetometer is sensitive towards ferrous metals in its immediate surroundings. This temporarily affects the sensor's ability to detect magnetic north, but may be used to detect the passage through a steel door frame. This method may not yield unique results, but if combined with a particle filter, may help to eliminate a few solutions and expedite convergence. 3.2.6.2.QR-Codes The so-called quick response codes or QR-codes are now seen everywhere on billboards and posters. There they are mostly used to encode a URL. The user just has to scan the code using his internet-enabled smartphone and is directed to the website for further information. This technique may also be used to transmit the position of the code to the smartphone. This can be achieved either by a unique identifier which is transmitted to a server, matched to a database and the position sent back to the smartphone, or by hard-coding the position data into the code itself. The accuracy is depending on the distance of the smartphone to the QR-code and the accuracy inherent to the method used to measure the QR-code's position upon set-up. An example code can be seen in Fig. 3. 3.2.6.3.Line recognition, pattern recognition, scene analysis The smartphone's camera provides a continuous, high-resolution image and can be used for image analysis. This can start from simple line detection and the consecutive recognition of hallways and intersections. Pattern recognition algorithms from the computer vision field of research may be able to identify objects such as doors, windows and stairways. Finally, camera input can be used in a complex scene analysis of a part of the building to provide orientation and position information. The limiting factor here is again the smartphone's system resources. 3.3 Smartphone Platform Upon selecting a smartphone platform for the project, the decision fell in favour of the N900 (Fig. 4) smartphone by Nokia. The N900 is a Linux-based phone with touch screen and sliding keyboard. It runs the open Maemo 5 operating system which is a Debian-based distribution. Contrary to Google Android-based phones, which are also Linux-based, Maemo 5 can run almost any Linux application as long as it has been cross compiled for the ARM Cortex architecture. As there is no complex virtualization and abstraction layer between the hardware and the application (as is with Android), applications developed for Maemo 5 can gain direct and low-level access to any attached (or internal) piece of hardware. This feature is essential for Stepping – smartphone-based portable pedestrian indoor navigation 320 Fig. 5 Custom INS - Linear accelerometer and gyroscope form the basis of the INS. They are supported by a barometric pressure sensor and a magnetometer. All sensors are integrated using a single digital signal processor and digital interfaces. Each image is being geo-referenced and the additional information scaled and orientated inside it according to the respective coordinates taken from a database. In that manner, the reality as taken by the picture is augmented by the additional information. 4. APPLICATIONS Generally, any application requiring the real-time knowledge of a person indoors may make use of the proposed system. A few cases are introduced here: 4.1 Public buildings and large plants On large plants business travellers often lose their way between production halls and office buildings. Using this system, however, they will not only find their way to the right building, but also to the right office inside that building, once they have crossed its threshold. This problem exists as well at large airports or public buildings such as libraries or universities. Assisted by the existing WiFi infrastructure the system will guide quickly to the destination counter, bookshelf or gate. Christian Lukianto, Harald Sternberg 321 4.2 Trade-fair or Congress navigation Visitors of large fairs or congresses usually have a limited amount of time to spend making contacts or negotiating business deals. For that reason they plan their visit beforehand and pick the booth identifiers from the exhibitor catalogue. However, the halls are huge and routes through them unclear. The problem is here one of efficient routing to the selected booths as well as between the different halls. This system, in combination with pre-installed navigation infrastructure in the buildings will address that problem. Fig. 6 PCB with mounted sensors, DSP and peripheral devices: Accelerometer, Pressure sensor, compass and gyroscope (bottom, left to right), Bluetooth module (centre-left, blue), serial port (lower right corner), USB and power (wiring on right hand side), DSP 4.3 Ambient Assisted Living Demographic studies show an ever increasing age of the members of our western civilization. Especially elderly citizens do not want to give up their independence by moving into retirement homes, surrender their driver's licenses or having to rely on other people. They want to live in normal dwellings as long as their mental and physical capabilities allow them to. Ambient Assisted Living (AAL) is a concept, which describes the combination of smart appliances and networks within the building to aid its inhabitants with day-to-day tasks. This system is ideal for AAL-enabled homes as it does not hinder the user's mobility but may even in cases of emergency be used as an alarm device to call for help. Stepping – smartphone-based portable pedestrian indoor navigation 322 5. CURRENT PROJECT STATE Currently, development efforts focus on the development of the custom INS and the evaluation of the smartphone's internal devices. 5.1 Custom INS The INS consists solely of digital MEMS sensors which are integrated using a single digital signal processor (DSP), as is outlined in Fig. 5. The linear accelerometer and the gyroscope sensor form the basis of the INS as their information is continuously processed in a strapdown algorithm (Savage, 2007). The magnetometer is used during the initialization stage and to help reduce gyro drift. The barometric pressure sensor is used in a feedback loop to control the vertical channel, i.e. the altitude computation of the strapdown algorithm. As the INS is a feasibility study to evaluate MEMS sensor capabilities under pedestrian navigation conditions, multiple communication interfaces on the INS are being evaluated: There is a serial interface, a USB interface and a Bluetooth transceiver on the INS, of which the best will be kept during the next design stage. The current INS PCB with mounted sensors and DSP is shown in Fig. 6. 5.2 Smartphone Internal Devices So far, the internal accelerometer and the GPS receiver have been examined. The accelerometer chip is connected via an I2C interface to the smartphone's central processor. Its readings can be accessed directly through the Linux sysfs-interface, or by using the Qt-Mobility API. Readings are available at rates up to 250Hz. The GPS receiver is not available as a separate module. Instead, it is connected to the 3G-communication module, resulting in a combined GPS/3G positioning module which is accessible through the Qt-Mobility API. The advantage here is, that GPS positioning and cell-based positioning, as well as their combination: the combined position measurement is available without much programming effort. The API allows for a selection of either or both positioning methods. 6. CONCLUSION AND OUTLOOK The proposed system presents a robust approach to the indoor navigation problem by making use of the ever increasing power of today's smartphones and the increasing quality of MEMS-based inertial sensors. Robustness with regard to missing external inputs is achieved by the modular nature of the implemented sensor fusion algorithm and the custom inertial navigation system connected to the smartphone. As this is a work in progress, further effort is to be invested on the fusion algorithm itself and the development and evaluation of the whole system in combination with the various external input types and consequently their respective subfilters. Furthermore, the INS will be improved so it can be used in a test environment outside of laboratory conditions. Christian Lukianto, Harald Sternberg