Jared B. Bancroft

Data Fusion Algorithms for Multiple Inertial Measurement Units

A single low cost inertial measurement unit (IMU) is often used in conjunction with GPS to increase the accuracy and improve the availability of the navigation solution for a pedestrian navigation system. This paper develops several fusion algorithms for using multiple IMUs to enhance performance. In particular, this research seeks to understand the benefits and detriments of each fusion method in the context of pedestrian navigation. Three fusion methods are proposed. First, all raw IMU measurements are mapped onto a common frame (i.e., a virtual frame) and processed in a typical combined GPS-IMU Kalman filter. Second, a large stacked filter is constructed of several IMUs. This filter construction allows for relative information between the IMUs to be used as updates. Third, a federated filter is used to process each IMU as a local filter. The output of each local filter is shared with a master filter, which in turn, shares information back with the local filters. The construction of each filter is discussed and improvements are made to the virtual IMU (VIMU) architecture, which is the most commonly used architecture in the literature. Since accuracy and availability are the most important characteristics of a pedestrian navigation system, the analysis of each filter’s performance focuses on these two parameters. Data was collected in two environments, one where GPS signals are moderately attenuated and another where signals are severely attenuated. Accuracy is shown as a function of architecture and the number of IMUs used.

Design and Testing of a Multi-Sensor Pedestrian Location and Navigation Platform

Navigation and location technologies are continually advancing, allowing ever higher accuracies and operation under ever more challenging conditions. The development of such technologies requires the rapid evaluation of a large number of sensors and related utilization strategies. The integration of Global Navigation Satellite Systems (GNSSs) such as the Global Positioning System (GPS) with accelerometers, gyros, barometers, magnetometers and other sensors is allowing for novel applications, but is hindered by the difficulties to test and compare integrated solutions using multiple sensor sets. In order to achieve compatibility and flexibility in terms of multiple sensors, an advanced adaptable platform is required. This paper describes the design and testing of the NavCube, a multi-sensor navigation, location and timing platform. The system provides a research tool for pedestrian navigation, location and body motion analysis in an unobtrusive form factor that enables in situ data collections with minimal gait and posture impact. Testing and examples of applications of the NavCube are provided.

Estimating MEMS gyroscope g-sensitivity errors in foot mounted navigation

Errors in gyroscope measurements due to linear accelerations are often overlooked in foot mounted navigation systems. Accelerations of foot mounted IMUs can reach 5 g while walking and 10 g while running, but vary depending on the sensors location and mounting. These accelerations are often very short and can induce large biases in the gyro which can produce attitude errors when the measurements are integrated. This paper proposes a real time method for the mitigation of g-sensitivity errors whereby the coefficients are estimated in the navigation Kalman filter. Variations of the estimation scheme are given including estimating the diagonal terms of the 3×3 matrix or all nine elements of the matrix. Accuracy (RMS) improved by 45% and 61% in two data sets using two different sensors in different environments. Convergence rates of the estimated variance are also shown.

Effect of camera characteristics on the accuracy of a visual gyroscope for indoor pedestrian navigation

Accurate navigation in GNSS degraded environments, namely indoors and in urban canyons, is still an unsolved problem. Other radio positioning systems used for indoor navigation purposes, like WLAN, are dependent on an infrastructure. Alternatively, if the initial position of the user is known, it may be propagated using self-contained sensors, such as digital compasses, rate gyroscopes and accelerometers. However, rate gyroscopes used for measuring angular velocity suffer from drift errors, and need augmentation to provide accurate positioning. Visual aiding is a substantial method for augmenting angular velocity measurements because it suffers from errors of a different nature than those of rate gyroscopes. The concept of a “visual gyroscope” described in this paper is based on tracking the motion of vanishing points in consecutive images and translating the motion information into the rotation of the camera. The motion of the camera may be further transformed into the heading change of the user. This paper assesses different cameras and setup characteristics defining the accuracy and functioning of a visual gyroscope using three different cameras and two setups.

Mitigation of attitude and gyro errors through vision aiding

Accurate positioning of first responders, electronic monitoring, and military personnel is often critical in GNSS denied environments. In such environments, inertial navigation systems (INS) are typically the preferred tool to be used for navigation. However, the gyros suffer from errors including biases, scale factors and g-dependent errors being the most significant ones. In order to sustain an accurate navigation solution for long durations, the gyroscope errors have to be measured and mitigated. Ideally, this calibration is done in situ. The attitude obtained using visual information is independent of the errors affecting the gyroscope. Human-made environments are commonly full of straight and parallel lines found in orthogonal directions. Perspective projection mapping transforms three-dimensional scenes into two-dimensional images. The process maintains the straight lines but modifies their parallelism resulting in an apparent point intersection of the lines. This point is called the vanishing point. Lines in three orthogonal directions constitute three vanishing points. The vanishing point locations are dependent on the camera rotation, but not camera translation. By monitoring the motion of the vanishing point locations in consecutive images, the relative roll, heading and pitch attitudes may be obtained. The absolute attitude, known from some a priori knowledge of the building layout, is then used to update the inertial navigation filter. Over time the visual measurements mitigate the cumulative errors of the gyro bias, scale factor and g-dependent bias. The performance of the vision-aided INS based navigation approach is evaluated herein. A camera is attached to a backpack and foot of a user moving through typical pedestrian based environments. The case of a foot-mounted camera is unique because of the high accelerations experienced during the human gait. The visual-aiding correction is found to significantly improve the attitude accuracy, especially the heading. Using the body solution, namely the camera and INS attached to a backpack, the vision-aiding yielded a 93 % improvement in the heading error during evaluation tests. With a foot-mounted solution, namely the INS and camera attached to the ankle of the user, the horizontal position error decreased by 34 %.

Use of magnetic quasi static field (QSF) updates for pedestrian navigation

This paper assesses a novel method of using a quasi-static magnetic field to provide updates to the navigation (i.e. attitude) filter. The method is able to make use of magnetometer measurements in a perturbed magnetic field, under the condition that the fields magnitude remains constant for short periods of time. The fact that magnetometer measurements can still be used in perturbed environments makes this update significant in terms of incorporating the magnetometer measurements into a navigation solution. The QSF process requires a detection algorithm to first identify quasi-static fields and second to perform the update. Thus this paper applies the QSF algorithm in a navigation filter to assess its performance in GNSS degraded or denied environments. Data sets are used to assess QSF updates. These range from open athletic fields to deep indoors where GPS signals are denied. The environments vary in terms of soft iron effects. The data was collected with high grade miniature MEMS IMUs, a high sensitivity GPS receiver and a low cost magnetometer. An accurate reference solution is derived from a tactical grade IMU. For the backpack mounted IMU the application of QSF updates yielded a 56 % heading error improvement when used as a heading reference system. For a corresponding ankle mounted system the application of QSF updates yielded a 56 % improvement in the position error (RMS) when used as a pedestrian navigation system. The maximum error over a 45 minute GPS outage decreased from 208 m to 128 m. The updates do not significantly decrease the estimated gyro error state variances, indicating that it is more suited for gyros and magnetometers with a lower performance than those used herein.

A New Approach for Improving Reliability of Personal Navigation Devices under Harsh GNSS Signal Conditions

In natural and urban canyon environments, Global Navigation Satellite System (GNSS) signals suffer from various challenges such as signal multipath, limited or lack of signal availability and poor geometry. Inertial sensors are often employed to improve the solution continuity under poor GNSS signal quality and availability conditions. Various fault detection schemes have been proposed in the literature to detect and remove biased GNSS measurements to obtain a more reliable navigation solution. However, many of these methods are found to be sub-optimal and often lead to unavailability of reliability measures, mostly because of the improper characterization of the measurement errors. A robust filtering architecture is thus proposed which assumes a heavy-tailed distribution for the measurement errors. Moreover, the proposed filter is capable of adapting to the changing GNSS signal conditions such as when moving from open sky conditions to deep canyons. Results obtained by processing data collected in various GNSS challenged environments show that the proposed scheme provides a robust navigation solution without having to excessively reject usable measurements. The tests reported herein show improvements of nearly 15% and 80% for position accuracy and reliability, respectively, when applying the above approach.

Enhanced pedestrian attitude estimation using vision aiding

Inertial Navigation System (INS) sensors are widely used for augmenting Global Navigation Satellite System measurements in urban environments and in the indoors. With a known initial position, the current position may be propagated using gyroscopes and accelerometers forming the INS for a limited time. The limitation of the self-contained sensors is the cumulative measurement errors that affect the accuracy of the attitude obtained using the gyroscopes. Vision aiding has proven to be a feasible method for mitigating these errors. This paper introduces a method to obtain attitude measurements by tracking the motion of vanishing points in consecutive images and integrating these measurements with the attitude observed by INS using an extended Kalman filter. The experiments show that vision aiding results in significant improvement of the user attitude and therefore the navigation solution. The challenges in vanishing point-based vision aiding are the processing time and the methods lack of capability to perceive sharp turns. These issues are addressed by developing an algorithm based on the Probabilistic Hough Transform for more efficient vanishing point calculation which also provides a means for turn detection. These improvements advance the objective of developing a real-time seamless indoor–outdoor pedestrian navigation system utilising vision aiding.

Improving the reliability of personal navigation devices in harsh environments

With advances in microelectromechanical system (MEMS) technology, many modern personal navigation devices incorporate measurements from various low-cost sensors alongside Global Navigation Satellite Systems (GNSS) receivers. However, both GNSS and other low-cost sensors are prone to the occurrence of faults that are either un-modeled or poorly modeled. This affects the usability of such personal navigation devices in some applications where reliability is a critical parameter. This paper thus presents several algorithms to either detect and remove such faults or model them properly in order to improve performance, especially reliability, for low cost multi-sensor integrated navigation systems. The algorithms presented in this paper can be broadly categorized into two parts. The first part focuses on optimizing the use of GNSS measurements in harsh environments. This is done by replacing the assumption of normal distribution of GNSS measurements with that of a heavy-tailed distribution. Moreover, the covariance of such distribution is also adapted to match the true error characteristics of the surrounding environment with the aid of inertial units. The second part of the algorithm detects possible faults arising in various sensors. Based on the type of sensor fault, the algorithm either rejects some of the measurements before they enter the integration filter, issues a warning signal to indicate lack of reliability information or deems the navigation solution unusable. The analyses of the proposed algorithms showed that faults were detected successfully and that the performance of the navigation system was improved in terms of both reliability and accuracy.