Calibration Aspects of INS Navigation

P. Clausen / J. Skaloud (Dir.)

Lausanne, EPFL, 2019.

DOI : 10.5075/epfl-thesis-9357.

The use of a Bayesian filter (e.g., Kalman filter) for the fusion of information from satellite positioning and inertial navigation is a common approach in many applications, where the knowledge of position, velocity, and attitude in space are of great interest. The correctness of these estimates depends on many factors, among others the quality of the sensor measurements and the errors within, which are directly reflected in the filter design. A calibration process allows compensating for deterministic influences (which in return improve for instance qualitatively the attitude initialization) and their inherent stochastic error signals required for filtering. This thesis presents in the first part the development of methods to perform a thorough calibration of different sensors in-lab under controlled conditions and in-field for a simplified calibration with limited resources and equipment. The stochastic properties of error signals are analyzed in the second part. A novel approach called Generalized Method of Wavelet Moments (GMWM) allows investigating the error structure using wavelets, which is similar to the Allan variance. An intuitive online tool is presented, which grants simplified access to the GMWM framework that provides a consistent, identifiable, and computationally efficient estimation of stochastic model parameters. The parameters of these error models are then made dependent on an external covariate such as temperature or motion. Indeed, it is experimentally confirmed that these properties shape the stochastic behavior of the measurements and how the stochastic parameters relate functionally to the influence of the covariate. Later, such knowledge is included in the filter for the correct estimation of confidence levels. The successful implementation of these proposed concepts is validated in a fully functional drone-system for mapping purposes. A real-time calibration scheme is applied first in-lab, later in-field to initialize the navigation processor. Apart from the benefit of achieving considerably better estimates of the attitude, and in case of satellite signal outage also of the position, the calibration allows for a simplified fusion of redundant inertial sensors. The improved performances through calibration and sensor redundancy are attractive to drone mapping applications relying on an accurate direct or integrated orientation such as lightweight airborne laser scanning systems or frame-cameras, which are utilized in the experiments.


Airborne Hyperspectral Imaging of Lakes

K. S. Barbieux / B. Merminod; J. Skaloud (Dir.)

Lausanne, EPFL, 2018.

DOI : 10.5075/epfl-thesis-8543.

In a time of rising concern about climate change and pollution, the water quality of large lakes acts as an indicator of the health of the environment. To study the water quality at a large scale - up to several hundreds of kilometres - hyperspectral remote sensing is emerging as the main solution. Indeed, different quantities relevant to water quality, like turbidity or concentratrion in chlorophyll-a, can be measured using the spectral reflectance of the water column. Additionally, airborne and spaceborne sensors can cover large areas, thus allowing to study the water at a much larger scale than when simply taking water samples at specific points. Airborne hyperspectral imaging, in particular, offers an acceptable ground resolution - around a metre - which allows to map relevant quantities precisely. However, few existing projects deliver maps that have both a sufficient ground resolution and a large coverage. Furthermore, most existing sensors do not offer a fine spectral resolution, which is for instance crucial when studying the presence of chlorophyll-a, which can only be detected in a narrow range of the electromagnetic spectrum. This thesis presents our work with a hyperspectral sensor developed and used by the Geodetic Engineering Laboratory of EPFL in the Léman-Baïkal project, a cooperative work which aimed at studying both Lake Geneva (Switzerland) and Lake Baikal (Russia). The project included ultralight plane flights with an onboard pushbroom scanner, which allowed to collect data over large areas with a fine spectral resolution. Alongside the use of this sensor came problematics which are at the centre of this thesis: the georeferencing of the scan lines, their radiometric calibration, their analysis and the softwaremanagement of this data. In the following, we present a new method to georeference pushbroom scan lines that uses co-acquired frame images to perform coregistration and to achieve a georeferencing, which RMSE is up to 20 times smaller than the direct one. We propose an efficient radiometric self-calibration method to convert the sensor output to water-leaving reflectance; this method makes use of the visible peaks of atmospheric absorption to align the spectral bands with those of a reference acquisition, and uses the near infrared properties of deep water and vegetation to performabsolute calibration. The last part of the processing - the software management, including data compression - was solved by developing a software called HYPerspectral Orthorectification Software (HypOS). This software is the synthesis of our work, including the tools to performgeometric correction, radiometric calibration and data compression of our hyperspectral data. Two examples of applications are given: the first one deals with mapping chlorophyll-a in the Rhone Delta of Lake Geneva; the second, at a larger scale, uses satellite data to monitor ice coverage over large lakes like Onega or Ladoga (Russia).

Vehicle Dynamic Model Based Navigation for Small UAVs

M. Khaghani / J. Skaloud (Dir.)

EPFL, 2018.

DOI : 10.5075/epfl-thesis-8494.

The dominant navigation system for small civilian UAVs today is based on integration of inertial navigation system (INS) and global navigation satellite system (GNSS). This strategy works well to navigate the UAV, as long as proper reception of GNSS signal is maintained. However, when GNSS outage occurs, the INS-based navigation solution drifts very quickly, considering the limited quality of IMU(s) employed in INS for small UAVs. In beyond visual line of sight (BVLOS) flights, this poses the serious danger of losing the UAV and its eventual falling down. Limited payload capacity and cost for small UAVs, as well as the need for operating in different conditions, with limited visibility for example, make it challenging to find a solution to reach higher levels of navigation autonomy based on conventional approaches. This thesis aims to improve the accuracy of autonomous navigation for small UAVs by at least one order of magnitude. The proposed novel approach employs vehicle dynamic model (VDM) as process model within navigation system, and treats data from other sensors such as IMU, barometric altimeter, and GNSS receiver, whenever available, as observations within the system. Such improvement comes with extra effort required to determine the VDM parameters for any specific UAV. This work investigates the internal capability of the proposed system for estimating VDM parameters as part of the augmented state vector within an extended Kalman filter (EKF) as the estimator. This reduces the efforts required to setup such navigation system that is platform dependent. Multiple experimental flights using two custom made fixed-wing UAVs are presented together with Monte-Carlo simulations. The results reveal improvements of 1 to 2 orders of magnitude in navigation accuracy during GNSS outages of a few minutes' duration. Computational cost for the proposed VDM-based navigation does not exceed 3~times that of conventional INS-based systems, which establishes its applicability for online application. A global sensitivity analysis is presented, spotting the VDM parameters with higher influence on navigation performance. This provides insight for design of calibration procedures. The proposed VDM-based navigation system can be interesting for professional UAVs from at least two points of view. Firstly, it adds little to no extra hardware and cost to the UAV. Secondly and more importantly, it might be currently the only way to reach such significant improvement in navigation autonomy for small UAVs regardless of visibility conditions and electromagnetic signals reception. Possibly, such environmental condition independence for navigation system may be needed to obtain certifications from legal authorities to expand UAV applications to new types of mission.


Miniature hyperspectral systems

D. Constantin / B. Merminod; Y. Akhtman (Dir.)

Lausanne, EPFL, 2017.

DOI : 10.5075/epfl-thesis-7647.

Spectroscopy methods have been used for decades to obtain information about various materials, ranging from galaxies billions of light years from earth, to Petri dishes containing rich bacteria cultures. From spectrometers in chemical labs to spectral cameras on satellites, the high dimensionality of spectral data is proven to be an excellent source of information for detection and classification of materials. In this study, state-of-the-art solutions for aerial hyperspectral surveys are reviewed and improved upon. In the frame of Leman-Baikal project, stringent requirements stemming from large scale hyperspectral mapping, have led to the development of a novel pushbroom data acquisition system and automated processing pipeline. The resulting end-to-end solution was successfully deployed on ultralight aircraft and subsequently used to acquire and process more than 15 terabytes of spectral data over the course of three years. The spectral data collected has proved its usefulness in environmental monitoring, generating water turbidity maps and vegetation classifications. The insight gained from the pushbroom system experiments led to the design and prototyping of a compact snapshot hyperspectral system, well suited for unmanned aerial vehicles. Weighing only 250 g and being a frame camera, the snapshot system presents many advantages over the pushbroom, including compatibility with automated image stitching and georeferencing solutions. Similarly to the pushbroom platform, a camera has been prototyped and the corresponding processing pipeline has been implemented. However, being based on interferometric filters, the snapshot hyperspectral sensors require extremely accurate calibration due to their complex transmissions. Novel methods and devices are presented in this study to overcome the interferometric spectral transmission issue, using either machine learning or compressive sensing approaches. The snapshot hyperspectral solution has been used in multiple studies and is shown to be particularly useful in the case of precision agriculture, where important plant traits are shown to correlate strongly with computed spectral maps.

Integrated Sensor Orientation on Micro Aerial Vehicles

M. Rehak / J. Skaloud (Dir.)

Lausanne, EPFL, 2017.

DOI : 10.5075/epfl-thesis-7530.

Mapping with Micro Aerial Vehicles (MAVs whose weight does not exceed 5 kg) is gaining importance in applications, such as corridor mapping, road and pipeline inspections, or mapping of large areas with homogeneous surface structure, e.g. forest or agricultural fields. When cm-level accuracy is required, the classical approach of sensor orientation does not deliver satisfactory results unless a large number of ground control points (GCPs) is regularly distributed in the mapped area. This may not be a feasible method either due to the associated costs or terrain inaccessibility. This thesis addresses such issues by presenting a development of MAV platforms with navigation and imaging sensors that are able to perform integrated sensor orientation (ISO). This method combines image measurements with GNSS or GNSS/IMU (Global Navigation Satellite System/Inertial Measurement Unit) observations. This innovative approach allows mapping with cm-level accuracy without the support of GCPs, even in geometrically challenging scenarios, such as corridors. The presented solution also helps in situations where automatic image observations cannot be generated, e.g. over water, sand, or other surfaces with low variations of texture. The application of ISO to MAV photogrammetry is a novel solution and its implementation brings new engineering and research challenges due to a limited payload capacity and quality of employed sensors on-board. These challenges are addressed using traditional as well as novel methods of treating observations within the developed processing software. The capability of the constructed MAV platforms and processing tools is tested in real mapping scenarios. It is empirically confirmed that accurate aerial control combined with a state-of-the-art calibration and processing can deliver cm-level ground accuracy, even in the most demanding projects. This thesis also presents an innovative way of mission planning in challenging environments. Indeed, a thorough pre-flight analysis is important not only for obtaining satisfactory mapping quality, but photogrammetric missions must be carried out in compliance with state regulations.


Modeling and Processing Approaches for Integrated Inertial Navigation

Y. Stebler / J. Skaloud (Dir.)

Lausanne, EPFL, 2013.

DOI : 10.5075/epfl-thesis-5601.


Cell-Based Deformation Monitoring via 3D Point Clouds

J. Wu / B. Merminod (Dir.)

Lausanne, EPFL, 2012.

DOI : 10.5075/epfl-thesis-5399.

Deformation is one of the most important phenomena in environmental science and engineering. Deformation of artificial and natural objects happens worldwide, such as structural deformation, landslide, subsidence, erosion, and rockfall. Monitoring and assessment of such deformation process is not only scientifically interesting, but also beneficial to hazard/risk control and prediction. In addition, it is also useful for regional planning and development. Deformation monitoring was driven by geodetic observations in the field of traditional geodetic surveying, based on the measurement of sparse points in a control network. Recently, with the rapid development of terrestrial LiDAR techniques, millions of points with associated three-dimensional coordinates (known as "3D point clouds") can be promptly captured in a few minutes. Compared to traditional surveying, terrestrial LiDAR offers great potential for deformation monitoring, because of various advantages such as fast data capture, high data density, and precise 3D object representation. By analysing 3D point clouds, the objective of this thesis is to provide an effective and efficient approach for deformation monitoring. Towards this goal, this thesis designs a new concept of "deformation map" for deformation representation and a novel "cell-based approach" for deformation computation. The main outcome of this thesis is a novel and rich approach that is able to automatically and incrementally compute a deformation map that enables a better understanding of structural and natural hazards with heterogeneous deformation characteristics. This work includes several dedicated contributions as follows. Hybrid Deformation Modelling. This thesis firstly provides a comprehensive investigation on the modelling requirements of various deformation phenomena. The requirements concern three main aspects, i.e., what has deformation (deformation object), which type of deformation, and how to describe deformation. Based on this detailed requirement analysis, we propose a rich and hybrid deformation model. This model is composed of meta-deformation, sub-deformation and deformation map, corresponding to deformation for a small cell, for a partial area, and for the whole object, respectively. Cell-based Deformation Computation. In order to automatically and incrementally extract heterogeneous deformation of the whole monitored object, we bring the "cell" concept into deformation monitoring. This thesis builds a cell-based deformation computing framework, which consists of three key steps: split, detect, and merge. Split is to divide the space of the object into many cells (uniform or irregular); detect is to extract the meta-deformation for individual cells by analysing the inside point clouds at two epochs; and merge is to group adjacent cells with similar deformation together and to form a consistent sub-deformation. As the final result, an informative deformation map is computed for describing the deformation for the whole object. Evaluation of Cell-based Approach. To evaluate such hybrid modelling and cell-based deformation computation, this thesis extensively studies both synthetic and real-life point cloud datasets: (1) by imitating a landslide scenario, we generate synthetic data using Matlab programming and practical settings, and compare the cell-based approach with traditional non-cell based geodetic methods; (2) by analysing two real-life cases of deformation in Switzerland, we further validate our approach and compare the results with third party sources (e.g., results provided by a surveying company, results computed by using a commercial software like 3DReshaper). Extension of Cell-based Approach. At the last stages of this thesis work, we particularly focus on providing several technical extensions to enhance this cell-based deformation monitoring approach. The main extensions include: (1) supporting dynamic cells instead of uniform cells when splitting the entire object space, (2) finding cell correspondence for the deformation scenarios that have large deformation like rockfalls, (3) movement tracking with data-driven cells which have irregular cell shape that can be automatically determined by the deformation boundary itself, (4) designing an adaptive modelling strategy that is able to accordingly select a suitable model for detecting meta-deformation of cells, and (5) computing deformation evolution for a monitored object with more than two epochs of point cloud datasets.


In-flight quality assessment and data processing for airborne laser scanning

P. Schär / J. Skaloud (Dir.)

Lausanne, EPFL, 2010.

DOI : 10.5075/epfl-thesis-4590.

One of the main problems of today's Airborne Laser Scanning (ALS) systems is the lack of reliable data QA/QC (Quality Assurance/ Quality Control) within or shortly after the airborne survey campaign. This thesis presents the development of methods to perform automated processing and QA/QC of ALS data during data acquisition (in-flight). The backbone of these methods is an error propagation algorithm that estimates the expected accuracy of the point-cloud. The error propagation considers the uncertainties induced by the direct georeferencing (DG) and the changing scanning geometry. A novel methodology that derives the scanning geometry directly from the point-cloud and computes a final quality indicator for every laser point is developed. To predict the accuracy of the navigation solution, this research also proposes a methodology to estimate in realtime (RT) the likelihood of fixing the differential carrier-phase ambiguities during post-processing. Furthermore, an innovative procedure to describe the quality of digital terrain models (DTM) derived from ALS data is presented. The successful implementation of these concepts into a fully functional in-flight quality monitoring tool embedded in an ALS system is demonstrated. The proposed tool incorporates RT GPS/INS processing and point-cloud georeferencing. The general performance of the tool, the validity of the quality indicators and the achievable improvement in efficiency for ALS data acquisition are assessed in airborne surveys. It is shown that when using RTK GPS (real-time kinematics) as positioning mode, the tool can provide point-clouds with sub-decimeter accuracy in RT.


Hybridation MEMS/UWB pour la navigation pédestre intra-muros

V. J. T. Renaudin-Schouler / B. Merminod; M. Kasser (Dir.)

Lausanne, EPFL, 2009.

DOI : 10.5075/epfl-thesis-4429.

Facing the expansion of geolocation needs, illustrated by the GALILEO European project, the growth of Location Based Services (LBS) and the need to identify the location of emergency mobile phone calls in Europe (standard E112), the research on localization techniques is booming. This thesis focuses on indoor pedestrian navigation and investigates a localization solution based on micro-electromechanical systems (MEMS) and ultra-wideband waves (UWB). MEMS based localization estimates the current location from a previously determined position using on-board low-cost inertial embedded sensors. Unfortunately, the performances of these autonomous systems are affected by large errors (typical of these sensors). In fact standalone solutions drift rapidly with time. Impulse-Radio UWB (IR-UWB) Times Of Arrival (TOA) are often used for localization purposes. This network based technology uses sensor networks, mainly attached to the infrastructure of the building to estimate the location of the transmitter with decimetre accuracy in ideal scenarii. However the indoor environment is hostile for radio propagation. Full of artificial obstacles, electromagnetic waves are disturbed and radiolocation performances are reduced. Construction materials also affect the magnetic field used to estimate the pedestrian's walking direction. In this context, the hybridization of these two complementary and uncorrelated technologies is promising. The study of the movement pattern of a pedestrian walking indoors induces two main outcomes on localization techniques. Firstly, random pedestrian movements complicate MEMS signal processing. Secondly, when the transmitter is worn by the user, for example around the neck, IR-UWB that propagates through the human body can hardly contribute to the localization. Optimal data fusion filters that hybridize a large set of observations : Angles Of Arrival (AOA), Time Differences Of Arrival (TDOA), accelerations, angular velocities and magnetic field measurements are presented. The coupling of UWB and MEMS data relies on an Extended Kalman Filter (EKF) complemented with specific procedures. Loose integration and tight integration are considered. Outlier detection processes within the radio data enrich the EKF. The most remarkable process is based on the RANSAC paradigm and employs the physical constraints of the pedestrian's walk described by biomechanics. In some cases, it enables the processing of reflected radio signals. A user equipped with a MEMS module and an UWB transceiver walked in the premises of the EPFL, following nine independent paths, for a total length of 380 m. The benefit of the MEMS/UWB hybridization filters are evaluated based on this experiment. The tight integration outperforms the loose coupling and enables indoor pedestrian localization with a one metre accuracy.

Trajectory determination and analysis in sports by satellite and inertial navigation

A. Wägli / J. Skaloud (Dir.)

Lausanne, EPFL, 2009.

DOI : 10.5075/epfl-thesis-4288.

This research presents methods for performance analysis in sports through the integration of Global Positioning System (GPS) measurements with Inertial Navigation System (INS). The described approach focuses on strapdown inertial navigation using Micro-Electro-Mechanical System (MEMS) Inertial Measurement Units (IMU). A simple inertial error model is proposed and its relevance is proven by comparison to reference data. The concept is then extended to a setup employing several MEMS-IMUs in parallel. The performance of the system is validated with experiments in skiing and motorcycling. The position accuracy achieved with the integrated system varies from decimeter level with dual-frequency differential GPS (DGPS) to 0.7 m for low-cost, single-frequency DGPS. Unlike the position, the velocity accuracy (0.2 m/s) and orientation accuracy (1 – 2 deg) are almost insensitive to the choice of the receiver hardware. The orientation performance, however, is improved by 30 – 50% when integrating four MEMS-IMUs in skew-redundant configuration. Later part of this research introduces a methodology for trajectory comparison. It is shown that trajectories based on dual-frequency GPS positions can be directly modeled and compared using cubic spline smoothing, while those derived from single-frequency DGPS require additional filtering and matching.


Algorithms for map-aided autonomous indoor pedestrian positioning and navigation

I. Spassov / B. Merminod; M. Bierlaire (Dir.)

Lausanne, EPFL, 2007.

DOI : 10.5075/epfl-thesis-3961.

The personal positioning and navigation became a very challenging topic in our dynamic time. The urban canyons and particularly indoors represent the most difficult areas for personal navigation problematic. Problems like disturbed satellite signals make the positioning impossible indoors. Recently developed systems for indoor positioning do not assure the necessary positioning accuracy or are very expensive. Our concept stands for a fully autonomous positioning and navigation process. That is, a method that does not rely on the reception of external information, like satellite or terrestrial signals. Therefore, this research is based on the use of inertial measurements of the human walk and the map database which contains the graphic representation of the elements of the building, created by applying the link-node model. Using this reduced set of information the task is to develop methodology, based on the interaction of the data from both sources, to assure reliable positioning and navigation process. This research is divided in three parts. The first part consists in the development of a methodology for initial localization of the person indoors. The problem to solve is to localize the person in the building. Consider a person equipped with a system which contains set of inertial sensors and map database of the building. Speed, turn rate and barometric altitude are measured and time-stamped on each step of the person. A pre-processing phase uses these raw measurements in order to construct a polyline, thus representing user's trajectory. In the localization approach central place takes the association of the user's trajectory with the graph representation of the building, process known as map-matching. The solution is based on statistical method where the determination of the user's position is entirely represented by its probability density function (PDF) in the frame of Bayesian inference. Initial localization determines the edge of the graph occupied by the person. The second part aims at continuous localization, where user's position is estimated on every step. Besides the application of the classical map-matching techniques, two new methods are developed. Both rely on the similarity of the geometry of the trajectory and the elements of the graph. The first is based on the Bayesian inference, where the estimation is computed considering the walked distance and azimuth. The second method represents a new application of the Fréchet distance as degree of similarity between two polylines. The third part is pointed at the pedestrian guidance. Once the user's position is known it is easy to compute the path to his destination and to give him directions. The problem is to assure continuance of the process of navigation in the case when the person has lost his path. In that case the solution consists in either giving instructions to the user to go back on the path or computation of a new path from the actual position of the user to his destination. Based on that methodology, algorithms for initial localization, continuous localization, and guidance were created. Numerous tests with the participation of several persons have been provided in order to validate the algorithms and to show their performance, robustness and limits.

Mobile mapping en temps réel pour la saisie automatique d'axes routiers

H. Gontran / B. Merminod (Dir.)

Lausanne, EPFL, 2007.

DOI : 10.5075/epfl-thesis-3749.

The development of road telematics requires the management of ever-growing databases related to traffic fluidity, live consignment monitoring and vehicle fleet tracking, as to driver assistance. Such an effort relies on the tight synergy between navigation technology, telecommunication and geographic information, to enhance the maintenance and exploitation of the road network and, above all, to strengthen security. Consequently, an accurate knowledge of the road environment and topology is mandatory to implement applications of transport telematics. The early nineties experienced major advances in GPS/INS coupling and the market launch of affordable digital cameras. Thus, a considerable portion of road information is captured by vehicles equipped with such sensors, a technique known as "mobile mapping". The advantage of the kinematic collection of data – such as the pavement geometry, its surfacing quality and the positioning of road objects – lies in the much faster completion of the survey, hence an excellent cost effectiveness. However, the complexity of data georeferencing and the fusion of the results with video sequences require numerous hours of repetitive labor. Moreover, only the process completion reveals the correct recording of position measurements. Any further survey can only be decided a few days later. We propose to introduce the concept of "real time" in the field of mobile mapping. The determinist exploitation of the data captured during a kinematic survey aims at restricting human intervention in the sophisticated georeferencing process, while authorizing the dissemination of this technique outside well-informed communities. The other challenge of this thesis that lies in the automatic fusion of localization data with images, under tight time constraints. In these conditions, what are the tools and algorithms robust enough to ensure the quality control of the georeferencing of road objects? We intend to provide these concerns a pertinent answer, while demonstrating the validity of the concept via the automatic acquisition and interpretation of the road geometry.


Development of a robotic mobile mapping system by vision-aided inertial navigation

F. A. Bayoud / B. Merminod (Dir.)

Lausanne, EPFL, 2006.

DOI : 10.5075/epfl-thesis-3440.

Vision-based inertial-aided navigation is gaining ground due to its many potential applications. In previous decades, the integration of vision and inertial sensors was monopolised by the defence industry due to its complexity and unrealistic economic burden. After the technology advancement, high-quality hardware and computing power became reachable for the investigation and realisation of various applications. In this thesis, a mapping system by vision-aided inertial navigation was developed for areas where GNSS signals are unreachable, for example, indoors, tunnels, city canyons, forests, etc. In this framework, a methodology on the integration of vision and inertial sensors was presented, analysed and tested when the only available information at the beginning is a number of features with known location/coordinates (with no GNSS signals accessibility), thus employing the method of "SLAM: Simultaneous Localisation And Mapping". SLAM is a term used in the robotics community to describe the problem of mapping the environment and at the same time using this map to determine (or to help in determining) the location of the mapping device. In addition to this, a link between the robotics and geomatics community was established where briefly the similarities and differences were outlined in terms of handling the navigation and mapping problem. Albeit many differences, the goal is common: developing a "navigation and mapping system" that is not bounded to the limits imposed by the used sensors. Classically, terrestrial robotics SLAM is approached using LASER scanners to locate the robot relative to a structured environment and to map this environment at the same time. However, outdoors robotics SLAM is not feasible with LASER scanners alone due to the environment's roughness and absence of simple geometric features. Recently in the robotics community, the use of visual methods, integrated with inertial sensors, has gained an interest. These visual methods rely on one or more cameras (or video) and make use of a single Kalman Filter with a state vector containing the map and the robot coordinates. This concept introduces high non-linearity and complications to the filter, which then needs to run at high rates (more than 20 Hz) with simplified navigation and mapping models. In this study, SLAM is developed using the Geomatics Engineering approach. Two filters are used in parallel: the Least-Squares Adjustment (LSA) for feature coordinates determination and the Kalman Filter (KF) for navigation correction. For this, a mobile mapping system (independent of GPS) is introduced by employing two CCD cameras (one metre apart) and one IMU. Conceptually, the outputs of the LSA photogrammetric resection (position and orientation) are used as the external measurements for the inertial KF. The filtered position and orientation are subsequently employed in the Photogrammetric intersection to map the surrounding features that are used as control points for the resection in the next epoch. In this manner, the KF takes the form of a navigation only filter, with a state vector containing the corrections to the navigation parameters. This way, the mapping and localisation can be updated at low rates (1 to 2 Hz) and use more complete modelling. Results show that this method is feasible with limitation induced from the quality of the images and the number of used features. Although simulation showed that (depending on the image geometry) determining the features' coordinates with an accuracy of 5-10 cm for objects at distances of up to 10 metres is possible, in practice this is not achieved with the employed hardware and pixel measurement techniques. Navigational accuracies depend as well on the quality of the images and the number and accuracy of the points used in the resection. While more than 25 points are needed to achieve centimetre accuracy from resection, they have to be within a distance of 10 metres from the cameras; otherwise, the resulting resection output will be of insufficient accuracy and further integration quality deteriorates. The initial conditions highly affect SLAM performance; these are the method of IMU initialisation and the a-priori assumptions on error distribution. The geometry of the system will furthermore have a consequence on possible applications. To conclude, the development consisted in establishing a mathematical framework, as well as implementing methods and algorithms for a novel integration methodology between vision and inertial sensors. The implementation and validation of the software have presented the main challenges, and it can be considered the first of a kind where all components were developed from scratch, with no pre-existing modules. Finally, simulations and practical tests were carried out, from which initial conclusions and recommendations were drawn to build upon. It is the author's hope that this work will stimulate others to investigate further this interesting problem taking into account the conclusions and recommendations sketched herein.


Capteurs et algorithmes pour la localisation autonome en mode pédestre

Q. Ladetto / B. Merminod (Dir.)

Lausanne, EPFL, 2003.

DOI : 10.5075/epfl-thesis-2710.

The challenge of knowing one's position in a precise and reliable way, at any time, with and without reception of satellite signals, represents an area fairly explored for the navigation of vehicles. To widen this service to the pedestrians requires a different approach that adapts to the dynamics, to the speed and especially to the total freedom of movement of the people. The traditional approach implements a triad of accelerometers and gyroscopes, which signals are integrated to obtain the relative displacement. This concept is unfortunately not judicious for a low-cost system. The principal reason is that the speed of displacement of a person is lost in the sensor noise level. In order to take into account all these specificities, an occurential approach was developed, based upon a subset of sensors as well as physiological and biomechanical parameters of the walk. This research is divided into three main directions. The first area of interest consists in the determination of the physiological parameters necessary to quantify the speed of walk and the step length. While the agitation of the accelerometer signals is a good speedometer, the frequency of the steps improves the robustness of the models. The influence of the gender added to the great human diversity imply the normalisation of the various relations deduced. Many tests carried out under conditions of everyday life reveal that the variation of the stride length, especially with the slope, strongly depends on the physical training of the person as well as on the duration of the climb or descent. Characteristic pattern were identified to differentiate between the forward, backward and lateral movements. The various suggested models were then favourably tested with some blind people, whose walking rhythm strongly varies according to the degree of confidence they have towards the course. The second part directly relates to the multiple technologies integrated to build an autonomous three-dimensional Pedestrian Navigation Module (PNM). The knowledge of the terrestrial magnetic field and its orientation makes it possible to determine the azimuth of displacement of a person. The use of a gyroscope improves the reliability of the system and facilitates the detection of magnetic disturbances. More stable in the short term than the compass, it is therefore the optimal complement under such circumstances. The altimetric information is obtained by barometric measurements which, according to the required precision, can be differential. The implementation of a GPS receiver allows the absolute positioning simultaneously to the calibration of the different sensor parameters and physiological models. The third part describes the integration of the models and measurements as well as the characteristics and treatments specific to pedestrian navigation. An initialisation phase is presented to individualize the parameters of the walk and adapt them from the general model. Hence, thanks to the compass-gyroscope integration together with the detection of any movement, this allows an optimal determination and filtering of the azimuth that has little or no temporal degradation. The consideration of several phenomena specific to the displacements of the humans brings artificial intelligence in pedestrian navigation. The coupling of the various sources of measurements, the influence of their precision on the computed position as well as their implication on the PNM reliability are described and illustrated. More than 550 km covered in various circumstances by 31 people allowed to validate the presented approach while fixing its limits.

GPS/INS integration for pedestrian navigation

V. Gabaglio / B. Merminod (Dir.)

Lausanne, EPFL, 2003.

DOI : 10.5075/epfl-thesis-2704.

This research has been sponsored by the Centre Suisse d'Electronique et de Microtechnique (CSEM) in Neuchâtel, Switzerland. It introduces a system and the algorithms for Pedestrian Navigation using a combination of sensors. The main objective is to localise a pedestrian anywhere and at any moment. The equipments utilised to fulfill this requirement are a Global Navigation Satellite System (GNSS) receiver and inertial sensors, which are attached to the person and as such need to be portable. An overview of Pedestrian Navigation constitutes the first part of the document. This new domain is examined from four different views: applications, tools (or sensors), architecture of the system and finally environment in which the pedestrian is travelling. As part of this process, the "state of the art" situation is presented. The approach followed in order to aid pedestrian to navigate is based on the Dead Reckoning technique coupled with GNSS. Consequently, the resolution of the travelled "distance" is separated from the resolution of the orientation of the walk. For the computation of the distance, a new technique based upon accelerometers and GNSS has been developed and demonstrated. The accelerometers are not used as a classical pedometer (counter of the number of steps) and the technique is not based on the double integration to obtain successively speed and distance. Instead, signal processing allows, considering individual parameters, the walking speed to be obtained directly from the signal of the accelerometers. This process, as well as the manner to determine the individual parameters, is presented in detail. The development of the algorithms is based on research performed in both the navigation and the medical domains (mainly in physiology). The computation of the orientation is more classical. It is based on the measurements made by a gyroscope and a GNSS receiver. The particularity of this study is the use of a single gyroscope to determine the orientation of the walk instead of three for the classical technique of inertial navigation. The influence of body movement on the gyroscope output has been deeply examined to determine the most appropriate way to process the signal of the gyroscope. The feasibility of the use of a single gyro, in the context of pedestrian navigation, is demonstrated. The potential added value for introducing a magnetic compass in the system is also evaluated. Finally a centralised Kalman filter has been designed and tested to merge all the sensors outputs, to combine the distance and the orientation, to integrate the Dead Reckoning solution and the GNSS solutions and to estimate all the parameters in a close to real-time process. The efficiency of this filter is demonstrated through different tests.


Suivi de la qualité des données spatiales au cours de leur acquisition et de leurs traitements

M. Azouzi / B. Merminod (Dir.)

Lausanne, EPFL, 2000.

DOI : 10.5075/epfl-thesis-2036.