[time]
2020-09-09T09:13:41+02:00

[team_name]
Free-Walking

[team_institution]
Beijing University of Posts and Telecommunications; Institute of Computing Technology, Chinese Academy of Sciences

[logolink]


[system_name]
Free-Walking

[website]


[track]
4

[reference_person]
Haiyong Luo, Qu Wang

[email]
wangqu@ict.ac.cn

[description]
ABSTRACT
A foot-mounted pedestrian inertial navigation system that accurately tracks pedestrian position by using inertial measurement units (IMUs) embedded in device when pedestrian walking normally. However, the positioning accuracy is decreased under complex movements. In this work, we divide the commonly used indoor walking motion type into eight modes construct motion mode classifier based on stacked denoising autoencoder and temporal convolutional network with attention to recognize these pedestrian motion modes. Base on the walking mode classification, we optimize the threshold or parameters of strapdown inertial navigation and zero-velocity detection, and Kalman filter for each walking mode. The proposed method can more effectively distinguish the pedestrian’s walking mode, accurately detect the stationary phase and estimate the device attitude under various pedestrian movements. The results of multi-floor and mixed walking modes experiment show that the positioning errors of the proposed method are less than 2%.
Keywords: Indoor localization; inertial navigation; walking mode recognition
 1.   INTRODUCTION
To provide indoor location-based services, a large number of indoor positioning and tracking technologies have been applied in academia as well as in industry. However, these methods rely on pre-installed infrastructures or training fingerprints, exclusively for positioning purposes, which requires extensive time and high costs for installation and maintenance. Moreover, there are many kinds of applications which are hard to pre-install in certain infrastructures or collect training fingerprints, e.g., in dangerous or collapsed buildings, emergency situations, etc, (Fig. 1). Among all the methods, the inertial navigation is widely used in an unknown environment, due to its independence from pre-installed infrastructures or prep-collected data.
With the booming of micro-electro-mechanical system (MEMS) technology, MEMS-based inertial navigation systems (INS) play an important role for providing location-based services in arbitrary unfamiliar and unprepared indoor and outdoor environments. Due to the integrative nature of the INS, any small bias will accumulate and grow with time boundlessly, especially when using the low-cost MEMS sensors. a Zero-Velocity Update (ZUPT)-aided INS technique has ability to suppress navigation errors. 
Many existing foot-mounted pedestrian inertial navigation systems have achieved relatively high positioning accuracy when pedestrian walking normally. However, human walking is diverse and complicated, how to complete the 3D positioning and navigation task under unconventional gait is still a problem. In this paper, we propose a 3D pedestrian inertial navigation method (named, Free-Walking) based on complicated walking mode recognition. Specifically, a walking mode classifier using the stacked denoising autoencoder and temporal convolutional network with attention is structured to distinguish walking mode. 
 
Fig. 1 Applications of foot-mounted pedestrian inertial navigation.

2.   SYSTEM DESIGN
The Free-Walking system combines data pre-processing, motion mode recognition, INS mechanization, adaptive zero velocity detection, ZUPT-aided Kalman filter and altitude constraint, as shown in Fig. 2.
 

Fig. 2. System architecture of proposed pedestrian inertial navigation based on motion mode recognition.
2.1 Walking Mode Recognition
For pedestrian positioning, the human motion modes describe the overall movement of pedestrians. The pedestrian motion modes are particularly important for pedestrian navigation, while the pedestrian motion modes are variable during the procedure of pedestrian navigation. For instance, when a pedestrian pass through a very narrow path, his motion mode is lateral walking. To reach destination quickly, his motion mode is usually running. Therefore, we design a walking mode classifier (see Fig. 3) based on the stacked denoising autoencoder and temporal convolutional network with attention to recognize eight pedestrian motion modes: standing still, running, forward walking, backward walking, lateral walking, stairs walking, elevator and escalator taking. If the detected gait is walking, going backwards, walking sideways or running, the travel distance is calculated by strapdown inertial navigation algorithm. If the pedestrian is going upstairs or downstairs, the barometer data is applied to determine altitude under the constraint of stair height. Especially, the threshold of strapdown inertial navigation and zero-velocity detection for each walking model can be individually optimized.
 
 
Fig. 3. Pedestrian walking mode recognition based on the stacked denoising autoencoder and temporal convolutional network with attention.
 

2.2 INS Mechanization
With the accelerometer and gyroscope measurements in the body frame, i.e., f ̃^b={a_x,a_y,a_z}  and w ̃^b={w_x,w_y,w_z}, the INS mechanization is used to derive the pedestrian’s positions p_^n, velocity v_^n, and attitudea_^n in the navigation frame by integrating acceleration twice and angular rate once after knowing the initial navigation state (i.e., 3D position, 3D velocity, and 3D attitude), as shown in Fig. 2. Due to the low quality of MEMS IMU, the slight navigation performance enhanced by certain small error correction terms (i.e., rotation of the Earth) of the INS mechanization is neglected. The INS mechanization equations can be presented in simplified form as follows:
	(█(p_k^n@v_k^n@〖a﷩b,k〗_ ))=(█(p_(k-1)^n v_k^n ∆t@v_(k-1)^n [a_(b,k)^n (f ̃_k^b-b_a )-g^n]@a_(b,k-1)^n a_(b,k-1)^n Ω[(w ̃_k^b-b_g)∆t])∆t)	(1)

where k is a time index, dt is the time differential, p^n represents the position vector in the navigation coordinate system frame (i.e., n-frame); v^nis the velocity vector in the n-frame; a_b^n is the transformation matrix from the body coordinate system (i.e., b-frame) to the n-frame as a function of attitude components; g^n=〖[0,0,-g]〗^T and is the Earth gravity vector in the n-frame;  f ̃^b and  w ̃^b are the acceleration and angle rate measurement vector, respectively; b_a and b_g are the bias vector of the tri-accelerometer and tri-gyroscope, respectively; ∆t=t_k-t_(k-1) is the time interval between the (k - 1)-th and k-th epochs; and Ω[•] is the cross-product form.
2.3 Zero Velocity Detection based on Motion Mode and Periodicity
The majority of zero-velocity detection methods employ comparisons between thresholds and the magnitude of acceleration, moving variance of acceleration, magnitude of angular rate, or their combinations. The primary limitation of these methods is that the variations in acceleration and angular rate differ greatly under various movement modes, such as walking, running, stair-climbing, etc. Thus, it is difficult to find a threshold function or threshold value that is widely applicable. In an experiment, we found that the zero-speed points in each gait cycle always occur around the minimum value of the zero-speed interval detector. Base on that rule, we propose a novel stationary phase detection method, which is not based on the threshold comparison principle but based on foot motion periodicity. We use the periodic gait-cycle window to divide the pedestrian movement into discrete gait cycles; then, we take the minimum value in each gait cycle as the zero-speed state point. The time length of the gait cycle is different under different motions. We adaptively adjust the gait-cycle duration based on the classification result of walking mode to adapt to various pedestrian motions. Compared to the existing methods, the proposed method does not need to set the zero-speed detection threshold, and performs well for zero-speed interval detection under various pedestrian movements.
2.4 ZUPT-aided Kalman Filter
Due to the integrative nature of the INS, any small bias will accumulate and grow with time boundlessly, especially when using the low-cost MEMS sensors. Fortunately, a Zero-Velocity Update (ZUPT) -aided INS technique has ability to suppress navigation errors. We take advantage of the stationary state of the foot during the stance phase and feeds the zero-velocity information (pseudo-measurement) into Kalman Filter (KF) to compensate for the velocity, the position and the attitude errors.
2.5 Height Constraint based on Walking Mode
The height errors in SINS solution will grow without boundary and cannot be eliminated by ZUPT measurements. When a user walks on the same floor, the altitude does not change. Only the altitude changes when the user goes up and down stairs. Therefore, the vertical displacement of pedestrian is constrained by two factors: stair height and motion mode. If the height of each stair in a multi-floor building is fixed, the height of each gait cycle is determined by the number of walking stairs in that gait cycle. Therefore, we use the classification result of walking mode to constrain the height error.


[references]
[1].	Qu Wang, Haiyong Luo*, Hao Xiong, Aidong Men*, Fang Zhao, Ming Xia, and Changhai Ou. Pedestrian Dead Reckoning based on Walking Pattern Recognition and Online Magnetic Fingerprint Trajectory Calibration. IEEE Internet of Things Journal. Aug. 2020, Vol. pp,1-1. Doi： 10.1109/JIOT.2020.3016146
[2].	Qu Wang, Haiyong Luo*, Langlang Ye, Aidong Men, Fang Zhao, Yan Huang, and Changhai Ou. Personalized Stride-Length Estimation based on Online Learning. IEEE Internet of Things Journal. Feb. 2020, Vol. 7, pp, 4885-4897. Doi:10.1109/JIOT.2020.2971318.
[3].	Qu Wang, Langlang Ye, Haiyong Luo*, Aidong Men*, Fang Zhao, and Yan Huang. Pedestrian Stride-Length Estimation Based on LSTM and Denoising Autoencoders, Sensors, Feb. 2019, vol. 19, no. 4, p. 840. Doi: 10.3390/s19040840.
[4].	Xile. Gao, Haiyong. Luo, Qu. Wang, Fang. Zhao, Langlang Ye, and Yuexia Zhang. A Human Activity Recognition Algorithm Based on Stacking Denoising Autoencoder and LightGBM. Sensors, Feb. 2019, vol. 19, no. 4, p. 947. Doi: 10.3390/s19040947.