Instrumentation & Measurement Magazine 25-1 - 47

time-synchronized and spatially-aligned data. Moreover, it
should be supported by high-end ground-truth to enable reliable
research and development in this area.
This paper presents the authors' work to build an efficient
multisensor indoor pedestrian tracking system composed of
off-the-shelf, low-cost commercial sensors and connected embedded
platforms. The system is backed up by a high-end
time-synchronized 3D ground-truth system. The developed
system adopts client-server architecture where multiple embedded
platforms are reporting to a central server. The server
is dedicated to real-time synchronization, data visualization,
and storage management. The client embedded platforms
control wearable sensors. The entire system is implemented
and organized under the umbrella of the Robot Operating System
(ROS).
The dataset developed in this work will be released publicly
online upon the publication of the work. The aim is
to offer the dataset with the supporting ground-truth and
design/implementation tips/guidelines to the research community.
We discuss the hardware architecture of our logger
system, followed by a section dedicated to the software architecture.
Finally, experimental results are provided, followed
by conclusions.
Tracking Algorithms
The main tracking technologies utilized in the system are
shown in Fig. 1. First, the stereo-camera positional tracking solution
steps, depicted in Fig. 1a, are summarized as follows [8]:
features are extracted and matched between stereoviews and
frames; the features are triangulated using the camera stereo
pair; and finally, features are tracked between frames to estimate
3D transformations and optimize position estimates.
Second, the developed system used an inertial measurement
unit (IMU) to measure the forward motion by counting
the steps. The person's step length is then projected in the direction
of motion. This process is called Pedestrian Dead
Reckoning (PDR), which uses a peak detection algorithm to
detect steps [9]. Vertically projected acceleration data shows
strong peaks that correspond to human walking steps. Then
an Attitude and Heading Reference system (AHRS) projects
the user's step in the motion direction. The trajectory is built
by concatenating the estimated relative motion steps. The PDR
general block diagram is shown in Fig. 1d.
Third, a LiDAR scan can be projected as a point cloud. Scanmatching
algorithms can estimate a relative pose between two
scans and then finds the rigid body transformation (translation
+ rotation) that best aligns two point-cloud scans. The Iterative
Closest Point (ICP) approach [10]is one popular algorithm for
scanning shown in Fig. 1b. In the ICP algorithm, an error metric,
usually a distance from the source to the reference point
cloud, is minimized by iteratively revising the transformation.
Simultaneous Localization and Mapping (SLAM) is used
to generate drift-free localization ground-truth. Traditionally,
SLAM was solved using filtering [11]. However, recent research
has focused on non-linear least-squares optimization
[12]. Optimization-based approaches [13]used factor graphs
February 2022
to re-describe SLAM. A typical structure usually separates the
front-end and back-end of such SLAM systems. Graph SLAM
information flow is shown in Fig. 1c.
Finally, UWB Transceivers can be grouped into anchors
with known locations and tags with unknown locations to
estimate the tag pose. Once range measurement is available between
a tag and at least three anchors, the tag location can be
estimated using trilateration [14].
System Architecture
The proposed system comprises two main sub-systems: a Data
Acquisition and Processing System (DAPS) module and the
Remote-Controlled System (RCS) module. All of the sensors
are integrated on embedded platforms using the DAPS module,
while the RCS module monitors and controls the DAPS
module's status and operations. The integrated sensors include
a stereo vision camera system, multiple IMUs, a Global
Navigation Satellite System (GNSS) receiver, an UWB wireless
system (consists of a tag mobile receiver and multiple anchors)
in addition to a 3D LiDAR sensor. Fig. 2 shows the hardware
configuration of the developed system. The system incorporates
state-of-art sensors that span visual, inertial, laser, local
wireless networks, and global satellite positioning technologies.
The sensors are configured and controlled by multiple
computing platforms that communicate wirelessly and exchange
ROS messages. This section describes the individual
sensors and the computing platforms.
Sensors Overview
This subsection summarizes the specifications of the utilized
sensors, including the hardware and the logged sensory data.
Stereovision Tracking Module: We use the Intel Realsense T265
Tracking Stereovision camera module. This stereovision tracking
camera module outputs 3D pose data by processing inputs
from dual fisheye cameras and IMU. The IMU is a system-inpackage
for the detection of acceleration in three dimensions
and rotations in three dimensions. The fisheye images are used
in the process of producing 6 DoF data streamed to the host
platform.
Handheld 3D LiDAR Scanner System: We used a " ZEB-Revo "
portable 3D laser scanner from " GeoSLAM " that provides accurate
3D point cloud and pose data. As the user walks through
the area of interest, the data is captured on a body-mounted remote
terminal (RT) that transmits pose, point-cloud, raw IMU,
and raw LiDAR scans in real-time to the server. When the data
logging is concluded, the RT performs a post-processing trajectory
and point cloud optimization. In this stage, the system
detects any loops and reduces the overall drift by applying
loop closures. This post-processing optimization leads to a
high-end optimized trajectory that is used as ground truth in
our experiments.
Precise AHRS: To provide precise spatial alignment that enables
accurate transformation between different sensors,
IEEE Instrumentation & Measurement Magazine
47
Instrumentation & Measurement Magazine 25-1

Table of Contents for the Digital Edition of Instrumentation & Measurement Magazine 25-1
