Localization and velocity tracking of human via 3 IMU sensors

Highlights

• A velocity tracking and localization method using only three IMU sensors is introduced.

• The drift-free and accurate 3D velocity result is achieved.

• The location of the person is tracked based on this velocity result and the limb kinematic subsequently.

• The benchmark study shows that the velocity and localization accuracy is within 2% in both normal walking and jumping.

Abstract

In sports training and exercises like walking and jogging, the velocity and position of the exercise people is very crucial for motion evaluation. A simple wearable system and corresponding method for velocity monitoring using minimal sensors can be very useful for daily use. In this work, a velocity tracking and localization method using only three IMU sensors is introduced. The three sensors are located at the right shank, right thigh and the pelvis to measure the kinematics of the lower limbs. In the method, a reference root point on the pelvis is chosen to represent the velocity and location of the person. Through acceleration fine tuning algorithm, the acceleration data is refined and combined with the velocity calculated from body kinematics to get a drift-free and accurate 3D velocity result. The location of the person is tracked based on this velocity estimation and the limb kinematic subsequently. The benchmark study with the commercial optical reference shows that the error in velocity tracking is within 0.1 m/s and localization accuracy is within 2% in both normal walking, jogging and jumping. Due to the conveniences of the small-size wearable IMU sensors, this proposed velocity tracking and localization method is very useful in everyday exercises both indoor and outdoor.

Introduction

Tracking the human motion and the physical human–environment interaction via wearable sensors closely relies on the human body kinematics, kinetics, and the contact interactions with the environment. Among the large amount of parameters of human motion, the awareness of velocity and location of human is very important for applications like sports training, exercises and entertainment etc. In daily applications, the human subject usually acts in house or an open environment. Although additional infra-structure localization devices such as ultrasonic, radio frequency identification (RFID) and ultra-wideband (UWB) etc. can be installed in the surroundings for localization, in these sports and exercises related applications which take place in large areas, they are not economically viable. The GPS system is not precise enough for positioning accuracy below one meter and it is normally only available in open outdoor environments. In these daily applications, it is expected to use simple wearable sensors to track the location and velocity of the person without depending on external infra-structures. The wireless IMUs (inertial measurement units) wearable sensors are small and do not have the capture volume limitations. Therefore, they now show a great advantage to be widely applied in many daily motion tracking applications [1], [2], [3], [4]. Existing IMU based motion captures systems use many sensors (usually more than 10) to track the human motion data. In many applications where the full body motions are not concerned, methods and system with minimal sensors to track the key kinematic information of interests are necessary.

In sports training such as walking, heel-to-toe walking and jogging, monitoring the velocity is quite crucial to study the efficiency of the exerciser. The speed of a subject describes how fast and in which direction the person moves in space. Also, its integration provides an update of the location data. Therefore, tracking the human moving velocity using wearable sensor is very helpful for such applications.

Theoretically, root point velocity can be estimated by integrating the acceleration of the root point over a period of time. However, the slight acceleration errors [5] measured by the IMUS will lead to unbounded drifting errors in just a few seconds [6]. Thus, direct integration of the acceleration is not suitable for velocity tracking of daily applications. Simple model based methods [7] only provide an average velocity estimation and is not generally accurate for different walking speeds. In gait study and personal navigation, researchers use the foot-mounted IMU sensors to track the foot location and velocity [8], [9], [10], [11]. Among these methods, the ZUPTs algorithm is efficient for regular walking localization, where the foot velocity is also tracked while walking. Since the foot velocity is zero when it stance on the floor, it cannot provide the continuous body velocity during motion. Also, during dynamic behaviors with obvious contacts, the large acceleration error will increase the velocity tracking and localization error in this method.

Therefore, to better resolve this velocity tracking issue, new methods for the inertial sensor to track the general practical moving velocity and location is expected.

In the previous work, we developed an IMU sensors based motion capture system with velocity and location tracking capability [3], [4], [12], [13]. With calibrated skeleton model [14], contact phase detection, the 3D velocity, location and motions of the subject when the person walks, climbs up/down stairs etc. are tracked. The previous method is name as velocity based simultaneous localization and capture, V-SLAC. Eight IMUs were used in the early method.

In many application scenarios, people only concern about the trajectory and the moving speed of the subject. For example, in personal localization and jogging, the location and the velocity are the key parameters. In this case, it would be good if the system can have as few sensors as possible. Therefore, we started the work in this paper to improve the method through using only three IMUs for tracking.

In [3], the velocity of the root point is calculated based on the leg kinematics every time the right or the left foot contacts the ground. Therefore, wearable sensors on both legs are needed to track the kinematic parameters for the root velocity calculation. The velocity tracking during the non-contact phase comes from the integration of the root acceleration over time. Because of the root acceleration errors in the sensor measurement, the integration update on the velocity drifts gradually over time. Therefore, the reference velocity from the lower limb kinematics should show up frequently for every step in order to prevent large drifting errors in the velocity.

In this paper, the idea is that if the error of the root acceleration can be estimated and eliminated, the integrated root velocity result can be more reliable so that we can track only one leg's kinematics for the velocity update. In this case, the number of sensors for the tracking system will reduce. Thus, in this proposed method, the acceleration error is estimated based on calculating the drifting rate of the velocity. Subsequently, the acceleration error is fine tuned to eliminate the drifting effect.

The method is illustrated in Fig. 1. (1) Initially, the sensors and the limb lengths are calibrated. The initial position and velocity of the person is registered. (2) As the subject start to move, the shoe pad detect the right contact phase (sensors are mounted on right legs). The location and the velocity of the person are updated, and the acceleration of the subject is also updated as well. (3) When the right contact is not available, the velocity and location of the person are updated based on the integrations of the refined acceleration. (4) The whole capture cycle repeats. This method is name as acceleration based simultaneous localization and capture, A-SLAC.

The remaining parts of the paper are organized as follow: Section 2 introduces the velocity tracking and localization algorithm. Section 3 introduces the system devices. Section 4 shows the experimental results. Finally, Section 5 concludes this paper.