Kinematic Model-Based Pedestrian Dead Reckoning for Heading Correction and Lower Body Motion Tracking

Abstract

In this paper, we present a method for finding the enhanced heading and position of pedestrians by fusing the Zero velocity UPdaTe (ZUPT)-based pedestrian dead reckoning (PDR) and the kinematic constraints of the lower human body. ZUPT is a well known algorithm for PDR, and provides a sufficiently accurate position solution for short term periods, but it cannot guarantee a stable and reliable heading because it suffers from magnetic disturbance in determining heading angles, which degrades the overall position accuracy as time passes. The basic idea of the proposed algorithm is integrating the left and right foot positions obtained by ZUPTs with the heading and position information from an IMU mounted on the waist. To integrate this information, a kinematic model of the lower human body, which is calculated by using orientation sensors mounted on both thighs and calves, is adopted. We note that the position of the left and right feet cannot be apart because of the kinematic constraints of the body, so the kinematic model generates new measurements for the waist position. The Extended Kalman Filter (EKF) on the waist data that estimates and corrects error states uses these measurements and magnetic heading measurements, which enhances the heading accuracy. The updated position information is fed into the foot mounted sensors, and reupdate processes are performed to correct the position error of each foot. The proposed update-reupdate technique consequently ensures improved observability of error states and position accuracy. Moreover, the proposed method provides all the information about the lower human body, so that it can be applied more effectively to motion tracking. The effectiveness of the proposed algorithm is verified via experimental results, which show that a 1.25% Return Position Error (RPE) with respect to walking distance is achieved.

Pedestrian Navigation Method Based on Machine Learning and Gait Feature Assistance

“…In Equations (10) and 11, a kx , a ky , a kz represent the tri-axial acceleration or angular velocity collected by accelerometers or gyroscopes of virtual foot IMU respectively at moment k. n is the variance of the interval size; ε a 1 , ε a 2 are the thresholds set according to the sensor accuracy of the VIMU. λ 1 , λ 2 , λ 3 , λ 4 represent the detection results of zero-velocity interval under four different discriminant methods, respectively.…”

“…The PDR algorithm needs to conduct kinematic modeling of human body, calculate the step length according to parameters such as step frequency and leg length, and obtain the 3D pedestrian position information with the aid of magnetic sensors. However, the parameters of different individual models vary greatly and PDR theories also have limitations when dealing with complex gait types [9,10]. Strapdown inertial navigation methods assisted by ZUPT use accelerometers and gyroscopes to calculate the navigation parameters of the human feet by a SINS algorithm, and ZUPT algorithm is used to suppress the accumulation of navigation errors when human feet are in static gait phases.…”

Enhanced Pedestrian Navigation Based on Course Angle Error Estimation Using Cascaded Kalman Filters

“…The inertial navigation system (INS) approach, which uses an extended Kalman filter (EKF) and zero velocity updates (ZUPT) for estimating navigation and sensor error states, has also been applied for indoor pedestrian navigation applications [ 13 , 14 , 15 , 16 , 17 ], and many other applications recently [ 18 , 19 , 20 , 21 ]. This approach, which is generally called the INS-EKF-ZUPT approach, uses an inertial measurement unit (IMU) installed on a foot or shoe, and the characteristics of the walking gait cycle, which is composed of stance and swing phases.…”

“…In order to improve the heading and position accuracy, a magnetic sensor installed in the IMU can be used. Because the magnetic heading measurements are used for estimating heading error and yaw gyro bias, the position accuracy can be improved in general compared with the simple step-length estimation approach [ 13 , 15 ]. Although the magnetic sensors can provide seamless heading information, they can be easily contaminated by external magnetic sources or magnetic substances.…”

“…Although they can provide a reasonable solution for most situations, they cannot handle the exceptional cases in which pedestrians do not walk in general walking patterns. Multiple sensor fusion approaches using the kinematic skeleton model of the human body have also been developed [ 15 , 38 ], which require many sensors for positioning, and therefore, impose excessive computational burden. In addition, research to improve the navigation performance of a low-cost integrated navigation system has been carried out in various fields, especially for unmanned vehicles [ 39 , 40 , 41 , 42 ].…”