Wearable-based Pedestrian Inertial Navigation with Constraints based on Biomechanical Models
Dina Bousdar Ahmed, Institute of Communications and Navigation, German Aerospace Center (DLR), Germany; Kai Metzger, Technical University of Munich, Germany
The market of wearable devices is exploiting. The trend is that wearable devices are not only carried, like smart watches or smart glasses, but also worn, e.g. smart shirts. As usual, the goal is to increase the offer of services. In our case, we focus on localization and the services that can be offered through it. One example is the protection of vulnerable road users (VRU) in urban scenarios. This service requires estimating the position of the VRU, namely the pedestrian or the cyclist.
We are particularly interested in the application possibilities of inertial sensors. Depending on the body location of the inertial measurement unit (IMU), the strapdown algorithm or the step-length-and-heading estimation algorithm is implemented.
The main challenge of inertial navigation systems (INSs) is the drift in the heading. The latter is the error in the heading, which is caused mainly by the bias of the gyroscopes. The community addresses this challenge by combining inertial technology with other technologies like GPS, GNSS, WiFi, etc. Some works also combine two IMUs attached to, for example, each foot of the pedestrian.
The significance of this work lies on the fact that pedestrian INSs are a feasible low-cost solution for multiple applications. The drift in the heading, however, is a challenge that hinders the use of INSs. Although this challenge can be addressed by combining inertial technology with others, e.g. GPS or WiFi, such infrastructure is not always available. For example, GPS signals are not available indoors or in urban canyons. WiFi signals are, depending on the location of the access points, not strong enough.
Our work provides an approach that does not require additional technology. Only the understanding of human biomechanics is necessary, and this field is vastly studied in medicine.
We want to approach the challenge of the heading drift by understanding human biomechanics. In fact, we want to exploit that body-mounted inertial sensors provide information about the pedestrian's motion. This information can be incorporated within the INS to improve the position estimation of the pedestrian.
The goal of this paper is to analyze how information about human biomechanical motion can improve the position estimation of an INS. We use two IMUs attached to the pedestrian's thigh and foot, respectively.
We develop our work in three steps. Firstly, we provide a biomechanical model of the human leg to estimate the position of the joints, i.e. knee, ankle and metatarsal. The joints are connected by links, namely thigh, shank and foot. Secondly, we analyze the joints' position estimated by the model when the orientation of the thigh-mounted INS is given as input. Finally, we propose modifications to a thigh-mounted INS based on the previous biomechanical analysis. The modifications estimate the biases of the roll and pitch angles computed by the pocket INS. The performance of the thigh-mounted INS with and without applying the biomechanical model is analyzed.
This work has two key innovative steps. The first one is to relate the orientation estimation of an INS to biomechanical behavior through a leg model. The latter allows for analyzing the errors in orientation from a different point of view. We classify human motion in either coherent or incoherent. Coherent motions are those that can be performed naturally by the human body, e.g. bending the knee 90°. Incoherent motions are those that cannot be performed because of physical limitations of the human body, e.g. rotating the foot 180°.
The orientation of the thigh IMU is used to estimate the position of the metatarsal. For that purpose, the orientation of the thigh IMU is input to the model as the orientation of the hip. Since the hip and thigh are linked directly, we assume that their orientation is the same. Then, the orientations of the knee and ankle are approximated during the stance phase, which is detected thanks to the foot IMU. Finally, the expected metatarsal's position can be estimated during the foot's stance phase.
The estimated position of the metatarsal is compared to the expected biomechanical behaviour. The latter is known from medicine studies. The comparison reveals whether the orientation of the thigh IMU is coherent or incoherent with respect to human motion.
The second innovative step is the estimation, based on the expected human biomechanics, of corrections for the orientation of the pocket INS. These corrections are intended to make the orientation estimation of an INS coherent with the motion expected from a human while walking.
The corrections of the thigh IMU are estimated during the detected stance phases of the foot. Furthermore, the corrections are only estimated if the thigh IMU is not coherent with the expected motion.
The first result of this work is that we can observe incoherent human motions in the orientation estimated by an INS. This result is achieved by the model of the human leg based on joints and links.
The second result of this work is that the position estimation of an INS can be improved with biomechanical constraints. These constraints are based on the analysis of human biomechanics performed previously.
The conclusion of our work is that human biomechanics aids inertial navigation by correcting the incoherent orientation estimations. The latter are equivalent to incoherent motions, which can be corrected to be coherent.