Wearable-based pedestrian localization through fusion of inertial sensor measurements

Kurzfassung

In pedestrian localization, fusion techniques have been exploited to address the weaknesses of different types of localization systems. Inertial localization is a particular case where the community has attempted to combine inertial technology with other technologies like satellite-based navigation or WiFi. Less explored are the approaches that combine multiple inertial sensors mounted on the human body. We refer to these approaches as multi-inertial measurement unit (IMU) localization systems. In this work, we want to study multi-IMU localization systems for pedestrian localization. The overall research objective of this thesis is pedestrian localization by means of inertial sensor fusion. We aim to determine the benefits, in distance accuracy, heading accuracy and height accuracy, of combining two inertial sensors. They will be mounted on the pocket and the foot of the same leg, respectively. The reason for choosing these body locations is that they have characteristic features during walking. Moreover, the fact that we consider the same leg will allow us to observe possible correlations between the measurements of the foot IMU and the pocket IMU during the walk. Our first step is the design of the evaluation methodology, which is inspired by the methodologies implemented in indoor localization competitions. The evaluation methodology allows us to quantitatively assess the performance of the two inertial localization systems based on a pocket IMU and a foot IMU. With our study, we identify the strengths and weaknesses of each of these systems, which helps us proposing the multi-IMU localization systems that we develop later on. Our proposed multi-IMU localization systems focus on three main challenges of inertial localization. The first challenge is related to the need for calibration of the parameters of step length models, which are used in non-foot-mounted inertial localization systems. The second challenge is the height drift that affects foot-mounted inertial systems. Finally, the last challenge is the heading drift, which is a challenge common to all inertial localization systems. The proposed calibration method automatically estimates the parameters of a step length model of the pocket inertial navigation system (INS). We explain how the optimal value of the parameters of the step length model are estimated, given the step length from the foot INS and the pitch amplitude of the pocket INS. We make two proposals: the first one calibrates only one parameter of the step length model and the second one calibrates both parameters. The proposed calibration method has the advantage of the automatization. That is, there is no need for an external operator to manually calibrate the step length model of the pocket INS. In addition, the automatic calibration can be carried out in real time, which is not possible with either of the state of the art calibration methods. %One of the main disadvantages of inertial-based localization is the drift in the heading, i.e. the error accumulation in the heading estimation. Next, we address the challenges related to the height drift and the heading drift. We study two different alternatives: a loose coupling system and a tight coupling system. The proposed loose coupling system combines the outputs of the foot INS and the pocket INS. The goal is to obtain an improved position estimation with respect to the single-IMU localization systems. The first contribution of this system is the development of an algorithm to determine how accurate the heading of the foot INS is with respect to the heading of the pocket INS. The second contribution of the loose coupling system is that it leverages the complementary strengths of the single-IMU localization systems regarding the height estimation. The height error of the loose coupling system outperforms the height error of the foot INS and the pocket INS in 75% and 87%, respectively. The last multi-IMU localization system we develop is the tight coupling system, which combines the raw measurements of the pocket IMU and the foot IMU. We develop a biomechanical model of the human leg which is used to analyze the typical motion, in terms of the roll and pitch, of the leg limbs. We compare the typical motion of the leg limbs to the one derived from the inertial roll and inertial pitch. In this way, we are able to observe the errors of an inertial localization system without the need of a reference trajectory. In order to characterize the heading from the biomechanical point of view, we compare the heading of the thigh to the heading of the foot. The advantage of this analysis is that we can observe incoherences in the relative heading between the two body limbs, which is equivalent to the relative heading between the pocket IMU and the foot IMU, respectively. The findings of the biomechanical study are then integrated in a tight coupling system. A highlight is that we avoid the use of hard constraints by modelling the roll and pitch angles of the pocket IMU and foot IMU as Gaussian distributions. With our proposal, it is possible to keep the behaviour of these angles coherent with respect to human motion. Regarding the relative heading, we introduce a pseudo-measurement update on the slope of the relative heading. The tight coupling system reduces its heading error in 70% and 72% with respect to the heading error of the pocket INS and the foot INS, respectively. Moreover, the height error of the tight coupling system is reduced in 87% and 75% with respect to the height error of the pocket INS and the foot INS, respectively. The evaluation of the proposed methods leads to interesting observations. For instance, all the inertial localization systems have approximately the same average distance error. In contrast, the tight coupling system outperforms all the other inertial localization systems regarding the heading estimation. One of the highlights of the loose coupling system is the reduction of the height error. This decrease is the result of sampling the height of the foot INS only when the user is walking the stairs. We close our work by stating that multi-IMU localization systems are more accurate than single-IMU localization systems. This accuracy is reflected in a considerable improvement of the heading error and the height error. Therefore, we recommend the use of multiple IMUs placed on different parts of the body to improve the accuracy of an inertial localization system.