Figure 1: Raw accelerometer signals of 3 axes (X,Y and Z) collected from a static device. The significant noise and data
drifting in the signal will lead to wrong results in trajectory reconstruction.
2.2 IMU Signals
Recent advances in MEMS technique bring possibil-
ity of producing small and light inertial navigation
systems. However, the main disadvantage of MEMS
devices is its low accuracy, which is indicated by bias
and noise in their measurements as elaborated in the
work of Woodman (Woodman, 2007) and illustrated
in Fig.1. During position tracking, the accelerometer
signals are integrated twice, and therefore the errors
grow even rapidly.
Some researchers pay attention to reducing errors
caused by IMU devices. Yang et al. proposes a
zero velocity compensation (ZVC) mechanism to re-
duce the accumulative errors of IMUs (Yang et al.,
2004). Pedley applies linear least squares optimiza-
tion to compute the recalibration parameters from the
available measurements (Pedley, 2013).
Some other methods adopt Kalman Filter com-
bined with computer vision as an assistant of IMU
to improve accuracy. For instance, a VISINAV algo-
rithm is presented to enable planetary landing, utiliz-
ing an extended Kalman filter (EKF) to reduce errors
(Mourikis et al., 2009). In an extended Kalman filter,
a state space model is applied to estimate the naviga-
tion states (Fredrikstad, 2016). However, the error-
state vector, which is estimated in advance, has a di-
rect impact on the result, that is, large deviation of
error-state estimation leads to poor results.
Most of those methods are not designed for low-
cost MEMS devices, which are commonly used in
mobile devices but produce large errors.
Consequently, in this paper, we put emphasis on
the trajectory reconstruction from low-cost MEMS
IMU. A quadrotor drone is taken as an example of
mobile devices. During the course, we design differ-
ent error models for different types of errors, so that
diverse errors can be eliminated in targeted ways.
Error Model
Reducing Noise
Eliminating Bias
1
Preprocess
Mobile Device
Reconstructed trajectory
Calibrated data
2
4
Integration
IMU data
Model Establishment
Error Elimination
Figure 2: The pipeline of our method: 1) Data collection
and preprocessing; 2) Error model establishment; 3) Error
elimination; 4) Integration and trajectory reconstruction.
3 IMU-BASED TRAJECTORY
RECONSTRUCTION
In this paper, we propose a trajectory reconstruction
method for mobile devices, utilizing the measurement
of IMU and other sensors. The pipeline of our method
is illustrated in Fig.2. It consists of four phases:
1. Sensor data collection and preprocess;
2. Error model establishment for the sensor data;
3. Error elimination on the basis of our error model;
4. Trajectory reconstruction through integration of
calibrated accelerometer data.
First, sensor data, mainly including the ac-
celerometer, gyroscope and ultrasonic signals, is col-
lected discretely from the target mobile device. Since
trajectory is reconstructed in the inertial frame while
IMU data is collected in the body frame (Lee et al.,
GRAPP 2018 - International Conference on Computer Graphics Theory and Applications
50