Estefania Munoz Diaz

Step detector and step length estimator for an inertial pocket navigation system

The use of smartphones for pedestrian positioning applications has become important in recent years. Algorithms for detecting steps and estimating their length can be found throughout the literature. In this paper we present new algorithms that increase the accuracy of the estimated step length and decrease the number of undetected steps. Unlike state of the art algorithms, that use the acceleration to detect steps and the frequency or the acceleration to estimate the step length, we propose the use of the opening angle of the pedestrians leg. An experiment with 18 volunteers was conducted in order to prove the relationship between the opening angle of the leg and the step length. Our results show that the estimation of the step length is robust against speed changes and the trajectorys estimated length has less than 1% error. Our step detector outperforms state of the art algorithms by presenting a higher step detection rate in challenging scenarios. In conclusion, it is possible and convenient to make use of the opening angle of the leg and its relationship with the step length.

Inertial Pocket Navigation System: Unaided 3D Positioning

Inertial navigation systems use dead-reckoning to estimate the pedestrians position. There are two types of pedestrian dead-reckoning, the strapdown algorithm and the step-and-heading approach. Unlike the strapdown algorithm, which consists of the double integration of the three orthogonal accelerometer readings, the step-and-heading approach lacks the vertical displacement estimation. We propose the first step-and-heading approach based on unaided inertial data solving 3D positioning. We present a step detector for steps up and down and a novel vertical displacement estimator. Our navigation system uses the sensor introduced in the front pocket of the trousers, a likely location of a smartphone. The proposed algorithms are based on the opening angle of the leg or pitch angle. We analyzed our step detector and compared it with the state-of-the-art, as well as our already proposed step length estimator. Lastly, we assessed our vertical displacement estimator in a real-world scenario. We found that our algorithms outperform the literature step and heading algorithms and solve 3D positioning using unaided inertial data. Additionally, we found that with the pitch angle, five activities are distinguishable: standing, sitting, walking, walking up stairs and walking down stairs. This information complements the pedestrian location and is of interest for applications, such as elderly care.Inertial navigation systems use dead-reckoning to estimate the pedestrians position. There are two types of pedestrian dead-reckoning, the strapdown algorithm and the step-and-heading approach. Unlike the strapdown algorithm, which consists of the double integration of the three orthogonal accelerometer readings, the step-and-heading approach lacks the vertical displacement estimation. We propose the first step-and-heading approach based on unaided inertial data solving 3D positioning. We present a step detector for steps up and down and a novel vertical displacement estimator. Our navigation system uses the sensor introduced in the front pocket of the trousers, a likely location of a smartphone. The proposed algorithms are based on the opening angle of the leg or pitch angle. We analyzed our step detector and compared it with the state-of-the-art, as well as our already proposed step length estimator. Lastly, we assessed our vertical displacement estimator in a real-world scenario. We found that our algorithms outperform the literature step and heading algorithms and solve 3D positioning using unaided inertial data. Additionally, we found that with the pitch angle, five activities are distinguishable: standing, sitting, walking, walking up stairs and walking down stairs. This information complements the pedestrian location and is of interest for applications, such as elderly care.

Standalone inertial pocket navigation system

Positioning applications became more important in recent years not only for security applications, but also for the mass market. Having a pedestrian navigation system embedded in a mobile phone is a realistic solution since it is equipped with low-cost sensors and the smartphone is located in a non-obstructive way. The location of the smartphone is important, since the position estimation process depends on it. Therefore, we propose to distinguish between pocket or bag, phoning, texting and swinging. We present a standalone inertial pocket navigation system based on an inertial measurement unit. For the computation of the orientation, we have developed an attitude estimator based on an unscented Kalman filter. The update stage has two different updates based on the acceleration and the magnetic field. Therefore, a zero acceleration detector, a magnetic disturbances detector and a static periods detector have been developed. The odometry in our navigation system is computed through an extended Kalman filter. The position is predicted with a movement model which is periodically updated through position corrections computed by the position computer. It comprises a step detector and a step length estimator based on the norm of the acceleration. The performance of our attitude estimator in comparison with the ground truth orientation is shown. The rest of the handheld positions are also tested for orientation. Likewise, we show pocket odometries of different users with the floor plan superimposed.

Optimal sampling frequency and bias error modeling for foot-mounted IMUs

The use of foot-mounted inertial measurement units (IMUs) has shown promising results in providing accurate human odometry as a component of accurate indoor pedestrian navigation. The specifications of these sensors, such as the sampling frequency have to meet requirements related to human motion. We investigate the lowest usable sampling frequency: To do so, we evaluate the frequency distribution of different human motion like crawling, jumping or walking in different scenarios such as escalators, lifts, on carpet or grass, and with different footwear. These measurements indicate that certain movement patterns, as for instance going downstairs, upstairs, running or jumping contain more high frequency components. When using only a low sampling rate this high frequency information is lost. Hence, it is important to identify the lowest usable sampling frequency and sample with it if possible. We have made a set of walks to illustrate the resulting odometries at different frequencies, after applying an Unscented Kalman Filter (UKF) using Zero Velocity Updates. The odometry error is highly dependent on the drift of the individual accelerometers and gyroscopes. In order to obtain better odometry it is necessary to perform a detailed analysis of the sensor noise processes. We resorted to computing the Allan variance for three different IMU chipsets of various quality specification. From this we have derived a bias model for the UKF and evaluated the benefit of applying this model to a set of real data from walk.

Bayesian cooperative relative vehicle positioning using pseudorange differences

Forward collision warning systems, lane change assistants or cooperative adaptive cruise control are examples of safety relevant applications that rely on accurate relative positioning between vehicles. Current solutions estimate the position of surrounding vehicles by measuring the distance with a RADAR sensor or a camera system. The perception range of these sensors can be extended by the exchange of GNSS information between the vehicles using an inter-vehicle communication link. In this paper we analyze two competing strategies against each other: the subtraction of the absolute positions estimated in each vehicle and the differentiation of GNSS pseudoranges. The aim of the later approach is to cancel out correlated errors in both receivers and, thus, achieve a better relative position estimate. The theoretical analysis is backed with Monte-Carlo simulations and empirical measurements in real world scenarios. Further on, two Bayesian approaches that make use of pseudorange differences are proposed. In a Kalman Filter pseudorange and Doppler measurements are used to estimate the baseline between two vehicles. This is extended in a second filter using on-board inertial and speed sensors following a multisensor fusion approach. The performance is evaluated in both, a highway and an urban scenario. The multisensor fusion approach proves to be able to stabilize the baseline estimate in GNSS challenging environments, like urban canyons and tunnels.

Bayesian recognition of safety relevant motion activities with inertial sensors and barometer

Activity recognition has been a hot topic in research throughout the last years. Walking, standing, sitting or lying have been detected with more or less confidence, in more or less suitable system designs. None of these systems however has entered daily life, neither in mass market, nor in professional environments. What is required is an unobtrusive system, requiring few resources and - most important - recognizing all important activities with high confidence. To this end, our research has focused on the professional market for safety related applications: first responders or also military use. Next to the classical motion related activities, our system supports motions in three dimensions that are necessary for all kinds of movements indoors as well as outdoors. These include falling, wriggling, crawling, climbing stairs up and down and using an elevator. We have proven this approach to run in real-time with only a single wireless sensor attached to the body while achieving robust and reliable recognition with a delay lower than two seconds.

Advanced pedestrian positioning system to smartphones and smartwatches

In recent years, there has been an increasing interest in the development of pedestrian navigation systems for satellite-denied scenarios. The popularization of smartphones and smartwatches is an interesting opportunity for reducing the infrastructure cost of the positioning systems. Nowadays, smartphones include inertial sensors that can be used in pedestrian dead-reckoning (PDR) algorithms for the estimation of the user’s position. Both smartphones and smartwatches include WiFi capabilities allowing the computation of the received signal strength (RSS). We develop a new method for the combination of RSS measurements from two different receivers using a Gaussian mixture model. We also analyze the implication of using a WiFi network designed for communication purposes in an indoor positioning system when the designer cannot control the network configuration. In this work, we design a hybrid positioning system that combines inertial measurements, from low-cost inertial sensors embedded in a smartphone, with RSS measurements through an extended Kalman filter. The system has been validated in a real scenario, and results show that our system improves the positioning accuracy of the PDR system thanks to the use of two WiFi receivers. The designed system obtains an accuracy up to 1.4 m in a scenario of 6000 m2.

Evaluation of AHRS algorithms for inertial personal localization in industrial environments

This paper presents a comparison among several state-of-the-art Attitude and Heading Reference Systems (AHRS). These algorithms can be used for 3D orientation and position estimation of users or devices. The robust performance of these AHRS algorithms is of paramount importance, specially in environments with potential external perturbations, such as industrial environments. The comparison among AHRS algorithms presented in this paper also includes an algorithm recently proposed by the authors (DLR-AHRS). In this paper the performance of the different AHRS will be studied, including the effect of magnetic perturbations on the performance of orientation estimation, and the effect of using different patterns of motion when the sensor is carried by a user at different locations (pocket, foot/shoe, hand). These AHRS algorithms are also compared with the Kalman-based commercially available AHRS algorithm of Xsens. The performance of the AHRS algorithms depends strongly on the strategies used to reject perturbations (sudden accelerations or deformations of the Earth magnetic field) and the ability of the systems to estimate the biases of the gyroscopes.

Performance comparison of foot- and pocket-mounted inertial navigation systems

The aim of this paper is to compare two inertial navigation systems in order to identify their strengths and weaknesses as well as to propose new fusion algorithms for these systems. The goal of the fusion is to combine the best of both navigation systems in order to obtain an improved position estimation of the pedestrian. To that extent, the comparison starts with an analysis of the sensor parameters obtained with the Allan variance method. The comparison continues with the analysis of the odometries obtained with both pocket- and foot-mounted sensors. This is done by comparing the true positions of a set of ground truth points with the position estimated by the navigation systems. The analysis shows that the pocket-mounted system is reliable for step detection, whereas the foot-mounted system misses some stance phase periods. The advantage of the foot-mounted system is, for example, the good estimation of the distance. For completeness and based on the results, we propose several fusion techniques to improve the pedestrians position estimation.

Use of the Magnetic Field for Improving Gyroscopes' Biases Estimation

An accurate orientation is crucial to a satisfactory position in pedestrian navigation. The orientation estimation, however, is greatly affected by errors like the biases of gyroscopes. In order to minimize the error in the orientation, the biases of gyroscopes must be estimated and subtracted. In the state of the art it has been proposed, but not proved, that the estimation of the biases can be accomplished using magnetic field measurements. The objective of this work is to evaluate the effectiveness of using magnetic field measurements to estimate the biases of medium-cost micro-electromechanical sensors (MEMS) gyroscopes. We carry out the evaluation with experiments that cover both, quasi-error-free turn rate and magnetic measurements and medium-cost MEMS turn rate and magnetic measurements. The impact of different homogeneous magnetic field distributions and magnetically perturbed environments is analyzed. Additionally, the effect of the successful biases subtraction on the orientation and the estimated trajectory is detailed. Our results show that the use of magnetic field measurements is beneficial to the correct biases estimation. Further, we show that different magnetic field distributions affect differently the biases estimation process. Moreover, the biases are likewise correctly estimated under perturbed magnetic fields. However, for indoor and urban scenarios the biases estimation process is very slow.An accurate orientation is crucial to a satisfactory position in pedestrian navigation. The orientation estimation, however, is greatly affected by errors like the biases of gyroscopes. In order to minimize the error in the orientation, the biases of gyroscopes must be estimated and subtracted. In the state of the art it has been proposed, but not proved, that the estimation of the biases can be accomplished using magnetic field measurements. The objective of this work is to evaluate the effectiveness of using magnetic field measurements to estimate the biases of medium-cost micro-electromechanical sensors (MEMS) gyroscopes. We carry out the evaluation with experiments that cover both, quasi-error-free turn rate and magnetic measurements and medium-cost MEMS turn rate and magnetic measurements. The impact of different homogeneous magnetic field distributions and magnetically perturbed environments is analyzed. Additionally, the effect of the successful biases subtraction on the orientation and the estimated trajectory is detailed. Our results show that the use of magnetic field measurements is beneficial to the correct biases estimation. Further, we show that different magnetic field distributions affect differently the biases estimation process. Moreover, the biases are likewise correctly estimated under perturbed magnetic fields. However, for indoor and urban scenarios the biases estimation process is very slow.