Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Σχεδιασμός και υλοποίηση αρθρωτού
συστήματος υπολογισμού τρισδιάστατης
θέσης κινούμενου αντικειμένου με
χρήση ετερογενών αισθητήρων και
φίλτρων
Διπλωματική Εργασία
Εκπόνηση:
Θεόδωρος Παπαφωτίου
AEM: 9708
Επίβλεψη:
Καθηγητής Ανδρέας Συμεωνίδης
Ερευνητής Εμμανουήλ Τσαρδούλιας
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Presentation Structure
2
Introduction
Methodology
Conclusions
Experiments
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Sensors calibration on
robots/devices by
using external local.
systems
Improve user
experience with more
accurate human
tracking
Problem Description
3
Why Indoor
Localization?
5
1
3
2
4
Navigation
Robot & human
navigation in
indoor spaces
Tours
Guided tours in
enclosed spaces
(i.e. museums)
LBS
Location-based
services in
airports, malls etc
Calibration
VR
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Thesis Purpose
4
01
02
03
04
Develop a complete, modular and user-friendly indoor
localization proposal, regardless the installation location
Integrate as many as possible indoor localization
methods and review them both individually and in
combinations
Develop custom software solutions for methods
implementations, fusion techniques and optimization
Develop custom design solutions for all systems
included in the project, that can be 3D-printed
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Indoor Localization Methods
5
● Particle Filter
● Kalman Filter
● Landmark
Detection
● Optical Flow
● Dead Reckoning
● Signal TX/RX
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Software
Architecture
Inputs provided by the systems’
equipment
Perception of the input data to
estimate the moving-device’s
3D-position & Filtering
Fusion of the estimated data
from each method based on
the selected methods from the
user
Output to the user via broker
and configuration
modifications to the systems’
equipment
6
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Software
Architecture
Inputs provided by the systems’
equipment
Perception of the input data to
estimate the moving-device’s
3D-position & Filtering
Fusion of the estimated data
from each method based on
the selected methods from the
user
Output to the user via broker
and configuration
modifications to the systems’
equipment
6
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Software
Architecture
Inputs provided by the systems’
equipment
Perception of the input data to
estimate the moving-device’s
3D-position & Filtering
Fusion of the estimated data
from each method based on
the selected methods from the
user
Output to the user via broker
and configuration
modifications to the systems’
equipment
6
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Software
Architecture
Inputs provided by the systems’
equipment
Perception of the input data to
estimate the moving-device’s
3D-position & Filtering
Fusion of the estimated data
from each method based on
the selected methods from the
user
Output to the user via broker
and configuration
modifications to the systems’
equipment
6
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Software
Architecture
Inputs provided by the systems’
equipment
Perception of the input data to
estimate the moving-device’s
3D-position & Filtering
Fusion of the estimated data
from each method based on
the selected methods from the
user
Output to the user via broker
and configuration
modifications to the systems’
equipment
6
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Localization Methods
Depth Frame Deprojecting
Fingerprinting
Dead reckoning
N-angulation
Optical Flow Calculation
7
LED pattern detection
Features Extraction & Tracking
Perspective Projection Model
9-DOF IMU data acquisition
UWB-nodes cluster RSSI
calculation
Camera Server
UWB Server
IMU Odometry
Visual Odometry
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Localization Methods
Depth Frame Deprojecting
Fingerprinting
Dead reckoning
N-angulation
Optical Flow Calculation
7
LED pattern detection
Features Extraction & Tracking
Perspective Projection Model
9-DOF IMU data acquisition
UWB-nodes cluster RSSI
calculation
Camera Server
UWB Server
IMU Odometry
Visual Odometry
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Localization Methods
Depth Frame Deprojecting
Fingerprinting
Dead reckoning
N-angulation
Optical Flow Calculation
7
LED pattern detection
Features Extraction & Tracking
Perspective Projection Model
9-DOF IMU data acquisition
UWB-nodes cluster RSSI
calculation
Camera Server
UWB Server
IMU Odometry
Visual Odometry
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Localization Methods
Depth Frame Deprojecting
Fingerprinting
Dead reckoning
N-angulation
Optical Flow Calculation
7
LED pattern detection
Features Extraction & Tracking
Perspective Projection Model
9-DOF IMU data acquisition
UWB-nodes cluster RSSI
calculation
Camera Server
UWB Server
IMU Odometry
Visual Odometry
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Fusion Methods
Lab Coordinates Outputs
of N Localization Methods
[ Xi Yi Zi vxi vyi vzi ], i = 1, 2 …
N
Kalman FIlter
k-Means
Weighted Average
Median
Selected Fusion Method by
user
Fused Estimation Output
[ Xf Yf Zf vxf vyf vzf ]
If N = 1, Fusion
process is skipped
We can implement a
“weighted” kalman filter by
adjusting its matrices
correspondingly
8
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Moving Localization Device
• LED pattern
• UWB tag
• ToF Ranging Sensor
• 9-DoF IMU
• PiCamera 3
• LCD screen
• Power Sensor
• RaspberryPi 4B
• Powerbank
X
Y
9
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Moving Device - IMU Odometry
X
Y
Calibrate IMU Sensors
(accelerometer/gyroscope/
magnetometer)
Acquire sensors data
(9x1 vector)
Insert data vector into a Kalman Filter
Retrieve Position, Velocity & Acceleration
of the IMU from Kalman Estimation
output
10
Relative
Position (!)
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Moving Device - Visual Odometry
Capture two sequential frames
[Ft], [Ft+1] and undistort them
Detect features on Ft
Track those features to Ft+1
Compute the Essential Matrix
Estimate
Rotation matrices &
Translation vectors
Extract X, Y, Z
World Coordinates
11
Relative
Position (!)
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Calculate relative camera positions
assuming their height known by
placing the device on pattern points
External System - Camera Server
Camera Server 1 Camera Server 2
Pan-Tilt Arm
RaspberryPi 4B
computational unit
D455 Intel RealSense
Color/Depth Camera
Rotate camera on Pan and Tilt to keep the tracked
device constantly inside the “active” area of the
frame
12
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Calculate relative camera positions
assuming their height known by
placing the device on pattern points
External System - Camera Server
Camera Server 1 Camera Server 2
Pan-Tilt Arm
RaspberryPi 4B
computational unit
D455 Intel RealSense
Color/Depth Camera
Rotate camera on Pan and Tilt to keep the tracked
device constantly inside the “active” area of the
frame
12
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
External System - Camera Server
Capture a frame
and undistort it
Separate the areas with HSV
values in a range, specified by
the tracked pattern color
Detect the contours in the
masked areas and select the
contour with the mostly
matching pattern shape
Change the frame
colorspace from
RGB to HSV
13
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Fingerprinting Perspective Projection Model
Depth Frame Point
Deprojection
External System - Camera Server
Multi-output Regression with
Support Vector Regression (SVR)
as the base model
14
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Fingerprinting Perspective Projection Model
Depth Frame Point
Deprojection
External System - Camera Server
Multi-output Regression with
Support Vector Regression (SVR)
as the base model
14
Rotation (!)
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
External System UWB-nodes
25
Tag
Anchor X
Initiator
Publish to
Broker
S1
S2 S3
S4
3D
Position
Estimation
Signal transmitted from
moving device UWB tag
N-angulation applied using
the Received Signal Strength
(RSS) from 4 anchors at each
timestep
16
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Graphical User Interface
Framework
for app dev.
Database for
data handling
17
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Experiments Environment
Laboratory of ECE AUTH
Research Student Team “VROOM”
● 2x Camera Servers placed diagonally at the area
● 6x UWB Anchors covering the area of interest
● 1x Computer for hosting the user app and the
database
18
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Experiments Environment
Points for moving experiments
Points for static experiments
19
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
All experiments
Moving experiments (2D) Static experiments
Total Displacement experiments
● Individual Methods
○ UWB system
○ Cameras system (Color)
○ Cameras system (Depth)
○ IMU Odometry
○ Visual Odometry
● Combinations
○ UWB & IMU Odometry
○ All Local methods
○ All Global methods
○ All methods
With & Without
Smoothing Filter
Only With
Smoothing Filter
● 2D movement (Ζ = ct)
○ UWB system & Cameras system (Color)
○ UWB system & Cameras system (Depth)
● 3D movement (Z = var)
○ UWB system & Cameras system (Color)
○ UWB system & Cameras system (Depth)
● All methods enabled - 2D movement
● All methods enabled - 3D movement
20
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Results - Moving Experiments
Without output filter With output filter
Kalman Filter &
Rolling median
21
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Results - Moving Experiments
Absolute Only Methods
All Methods
UWB System & IMU Odometry
22
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Conclusions
Absolute Localization (AL)
is more accurate compared to
Reference Localization (RL)
RL is highly affected by sharp alterations of the
moving direction
Camera Server localization methods provide
the best standalone localization results
The combination of all AL presents the best
performance through all experiments
The best low-cost combinations include the
UWB method standalone or its combination with
the IMU Odometry method
The Visual Odometry poses the greatest errors
due to the lack of enough visually trackable
features of the environment
23
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Results - Static Experiments
Z
=
ct
Z
=
var
UWB & CS-Color
24
UWB & CS-Depth
13cm
22cm
22cm
19cm
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Results - Static Experiments
Z
=
ct
Z
=
var
UWB & CS-Color
24
UWB & CS-Depth
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Future implications
Use of
BLE
transmitters for
additional low-
cost tracking
Simplification of
systems
installation at a
laboratory by
managing all the
labs through a
common broker,
database and
application that will
be hosted remotely
Custom
equipment
(PCB, connectors)
for the moving-
device to
significantly
reduce its size
Various patterns
on the moving
device to be
detected via the
Camera Server on
various device
orientation-angles
for optimal & robust
tracking
Fine-tuning of
localization methods
and fusion
techniques by
extensive testing
using robots, drones
and humans in
various indoor
environments
Integrate
Neural
Networks in the
Pattern
Detection, Noise
Filtering and
Fusion processes
Integration of
additional
localization
methods to
increase the range
of this project’s
applications
25
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Future implications
Use of
BLE
transmitters for
additional low-
cost tracking
Simplification of
systems
installation at a
laboratory by
managing all the
labs through a
common broker,
database and
application that will
be hosted remotely
Custom
equipment
(PCB, connectors)
for the moving-
device to
significantly
reduce its size
Various patterns
on the moving
device to be
detected via the
Camera Server on
various device
orientation-angles
for optimal & robust
tracking
Fine-tuning of
localization methods
and fusion
techniques by
extensive testing
using robots, drones
and humans in
various indoor
environments
Integrate
Neural
Networks in the
Pattern
Detection, Noise
Filtering and
Fusion processes
Integration of
additional
localization
methods to
increase the range
of this project’s
applications
25
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Future implications
Use of
BLE
transmitters for
additional low-
cost tracking
Simplification of
systems
installation at a
laboratory by
managing all the
labs through a
common broker,
database and
application that will
be hosted remotely
Custom
equipment
(PCB, connectors)
for the moving-
device to
significantly
reduce its size
Various patterns
on the moving
device to be
detected via the
Camera Server on
various device
orientation-angles
for optimal & robust
tracking
Fine-tuning of
localization methods
and fusion
techniques by
extensive testing
using robots, drones
and humans in
various indoor
environments
Integrate
Neural
Networks in the
Pattern
Detection, Noise
Filtering and
Fusion processes
Integration of
additional
localization
methods to
increase the range
of this project’s
applications
25
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Future implications
Use of
BLE
transmitters for
additional low-
cost tracking
Simplification of
systems
installation at a
laboratory by
managing all the
labs through a
common broker,
database and
application that will
be hosted remotely
Custom
equipment
(PCB, connectors)
for the moving-
device to
significantly
reduce its size
Various patterns
on the moving
device to be
detected via the
Camera Server on
various device
orientation-angles
for optimal & robust
tracking
Fine-tuning of
localization methods
and fusion
techniques by
extensive testing
using robots, drones
and humans in
various indoor
environments
Integrate
Neural
Networks in the
Pattern
Detection, Noise
Filtering and
Fusion processes
Integration of
additional
localization
methods to
increase the range
of this project’s
applications
25
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Future implications
Use of
BLE
transmitters for
additional low-
cost tracking
Simplification of
systems
installation at a
laboratory by
managing all the
labs through a
common broker,
database and
application that will
be hosted remotely
Custom
equipment
(PCB, connectors)
for the moving-
device to
significantly
reduce its size
Various patterns
on the moving
device to be
detected via the
Camera Server on
various device
orientation-angles
for optimal & robust
tracking
Fine-tuning of
localization methods
and fusion
techniques by
extensive testing
using robots, drones
and humans in
various indoor
environments
Integrate
Neural
Networks in the
Pattern
Detection, Noise
Filtering and
Fusion processes
Integration of
additional
localization
methods to
increase the range
of this project’s
applications
25
Design and implementation of an integrated and modular system for 3D position
calculation of a moving object using heterogeneous sensors and filters
01/04/2024
Thank you!
Questions?
26

Σχεδιασµός και υλοποίηση αρθρωτού συστήµατος υπολογισµού τρισδιάστατης θέσης κινούµενου αντικειµένου µε χρήση ετερογενών αισθητήρων και ϕίλτρων

  • 1.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Σχεδιασμός και υλοποίηση αρθρωτού συστήματος υπολογισμού τρισδιάστατης θέσης κινούμενου αντικειμένου με χρήση ετερογενών αισθητήρων και φίλτρων Διπλωματική Εργασία Εκπόνηση: Θεόδωρος Παπαφωτίου AEM: 9708 Επίβλεψη: Καθηγητής Ανδρέας Συμεωνίδης Ερευνητής Εμμανουήλ Τσαρδούλιας
  • 2.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Presentation Structure 2 Introduction Methodology Conclusions Experiments
  • 3.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Sensors calibration on robots/devices by using external local. systems Improve user experience with more accurate human tracking Problem Description 3 Why Indoor Localization? 5 1 3 2 4 Navigation Robot & human navigation in indoor spaces Tours Guided tours in enclosed spaces (i.e. museums) LBS Location-based services in airports, malls etc Calibration VR
  • 4.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Thesis Purpose 4 01 02 03 04 Develop a complete, modular and user-friendly indoor localization proposal, regardless the installation location Integrate as many as possible indoor localization methods and review them both individually and in combinations Develop custom software solutions for methods implementations, fusion techniques and optimization Develop custom design solutions for all systems included in the project, that can be 3D-printed
  • 5.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Indoor Localization Methods 5 ● Particle Filter ● Kalman Filter ● Landmark Detection ● Optical Flow ● Dead Reckoning ● Signal TX/RX
  • 6.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Software Architecture Inputs provided by the systems’ equipment Perception of the input data to estimate the moving-device’s 3D-position & Filtering Fusion of the estimated data from each method based on the selected methods from the user Output to the user via broker and configuration modifications to the systems’ equipment 6
  • 7.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Software Architecture Inputs provided by the systems’ equipment Perception of the input data to estimate the moving-device’s 3D-position & Filtering Fusion of the estimated data from each method based on the selected methods from the user Output to the user via broker and configuration modifications to the systems’ equipment 6
  • 8.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Software Architecture Inputs provided by the systems’ equipment Perception of the input data to estimate the moving-device’s 3D-position & Filtering Fusion of the estimated data from each method based on the selected methods from the user Output to the user via broker and configuration modifications to the systems’ equipment 6
  • 9.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Software Architecture Inputs provided by the systems’ equipment Perception of the input data to estimate the moving-device’s 3D-position & Filtering Fusion of the estimated data from each method based on the selected methods from the user Output to the user via broker and configuration modifications to the systems’ equipment 6
  • 10.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Software Architecture Inputs provided by the systems’ equipment Perception of the input data to estimate the moving-device’s 3D-position & Filtering Fusion of the estimated data from each method based on the selected methods from the user Output to the user via broker and configuration modifications to the systems’ equipment 6
  • 11.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Localization Methods Depth Frame Deprojecting Fingerprinting Dead reckoning N-angulation Optical Flow Calculation 7 LED pattern detection Features Extraction & Tracking Perspective Projection Model 9-DOF IMU data acquisition UWB-nodes cluster RSSI calculation Camera Server UWB Server IMU Odometry Visual Odometry
  • 12.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Localization Methods Depth Frame Deprojecting Fingerprinting Dead reckoning N-angulation Optical Flow Calculation 7 LED pattern detection Features Extraction & Tracking Perspective Projection Model 9-DOF IMU data acquisition UWB-nodes cluster RSSI calculation Camera Server UWB Server IMU Odometry Visual Odometry
  • 13.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Localization Methods Depth Frame Deprojecting Fingerprinting Dead reckoning N-angulation Optical Flow Calculation 7 LED pattern detection Features Extraction & Tracking Perspective Projection Model 9-DOF IMU data acquisition UWB-nodes cluster RSSI calculation Camera Server UWB Server IMU Odometry Visual Odometry
  • 14.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Localization Methods Depth Frame Deprojecting Fingerprinting Dead reckoning N-angulation Optical Flow Calculation 7 LED pattern detection Features Extraction & Tracking Perspective Projection Model 9-DOF IMU data acquisition UWB-nodes cluster RSSI calculation Camera Server UWB Server IMU Odometry Visual Odometry
  • 15.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Fusion Methods Lab Coordinates Outputs of N Localization Methods [ Xi Yi Zi vxi vyi vzi ], i = 1, 2 … N Kalman FIlter k-Means Weighted Average Median Selected Fusion Method by user Fused Estimation Output [ Xf Yf Zf vxf vyf vzf ] If N = 1, Fusion process is skipped We can implement a “weighted” kalman filter by adjusting its matrices correspondingly 8
  • 16.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Moving Localization Device • LED pattern • UWB tag • ToF Ranging Sensor • 9-DoF IMU • PiCamera 3 • LCD screen • Power Sensor • RaspberryPi 4B • Powerbank X Y 9
  • 17.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Moving Device - IMU Odometry X Y Calibrate IMU Sensors (accelerometer/gyroscope/ magnetometer) Acquire sensors data (9x1 vector) Insert data vector into a Kalman Filter Retrieve Position, Velocity & Acceleration of the IMU from Kalman Estimation output 10 Relative Position (!)
  • 18.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Moving Device - Visual Odometry Capture two sequential frames [Ft], [Ft+1] and undistort them Detect features on Ft Track those features to Ft+1 Compute the Essential Matrix Estimate Rotation matrices & Translation vectors Extract X, Y, Z World Coordinates 11 Relative Position (!)
  • 19.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Calculate relative camera positions assuming their height known by placing the device on pattern points External System - Camera Server Camera Server 1 Camera Server 2 Pan-Tilt Arm RaspberryPi 4B computational unit D455 Intel RealSense Color/Depth Camera Rotate camera on Pan and Tilt to keep the tracked device constantly inside the “active” area of the frame 12
  • 20.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Calculate relative camera positions assuming their height known by placing the device on pattern points External System - Camera Server Camera Server 1 Camera Server 2 Pan-Tilt Arm RaspberryPi 4B computational unit D455 Intel RealSense Color/Depth Camera Rotate camera on Pan and Tilt to keep the tracked device constantly inside the “active” area of the frame 12
  • 21.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 External System - Camera Server Capture a frame and undistort it Separate the areas with HSV values in a range, specified by the tracked pattern color Detect the contours in the masked areas and select the contour with the mostly matching pattern shape Change the frame colorspace from RGB to HSV 13
  • 22.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Fingerprinting Perspective Projection Model Depth Frame Point Deprojection External System - Camera Server Multi-output Regression with Support Vector Regression (SVR) as the base model 14
  • 23.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Fingerprinting Perspective Projection Model Depth Frame Point Deprojection External System - Camera Server Multi-output Regression with Support Vector Regression (SVR) as the base model 14 Rotation (!)
  • 24.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 External System UWB-nodes 25 Tag Anchor X Initiator Publish to Broker S1 S2 S3 S4 3D Position Estimation Signal transmitted from moving device UWB tag N-angulation applied using the Received Signal Strength (RSS) from 4 anchors at each timestep 16
  • 25.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Graphical User Interface Framework for app dev. Database for data handling 17
  • 26.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Experiments Environment Laboratory of ECE AUTH Research Student Team “VROOM” ● 2x Camera Servers placed diagonally at the area ● 6x UWB Anchors covering the area of interest ● 1x Computer for hosting the user app and the database 18
  • 27.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Experiments Environment Points for moving experiments Points for static experiments 19
  • 28.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 All experiments Moving experiments (2D) Static experiments Total Displacement experiments ● Individual Methods ○ UWB system ○ Cameras system (Color) ○ Cameras system (Depth) ○ IMU Odometry ○ Visual Odometry ● Combinations ○ UWB & IMU Odometry ○ All Local methods ○ All Global methods ○ All methods With & Without Smoothing Filter Only With Smoothing Filter ● 2D movement (Ζ = ct) ○ UWB system & Cameras system (Color) ○ UWB system & Cameras system (Depth) ● 3D movement (Z = var) ○ UWB system & Cameras system (Color) ○ UWB system & Cameras system (Depth) ● All methods enabled - 2D movement ● All methods enabled - 3D movement 20
  • 29.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Results - Moving Experiments Without output filter With output filter Kalman Filter & Rolling median 21
  • 30.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Results - Moving Experiments Absolute Only Methods All Methods UWB System & IMU Odometry 22
  • 31.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Conclusions Absolute Localization (AL) is more accurate compared to Reference Localization (RL) RL is highly affected by sharp alterations of the moving direction Camera Server localization methods provide the best standalone localization results The combination of all AL presents the best performance through all experiments The best low-cost combinations include the UWB method standalone or its combination with the IMU Odometry method The Visual Odometry poses the greatest errors due to the lack of enough visually trackable features of the environment 23
  • 32.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Results - Static Experiments Z = ct Z = var UWB & CS-Color 24 UWB & CS-Depth 13cm 22cm 22cm 19cm
  • 33.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Results - Static Experiments Z = ct Z = var UWB & CS-Color 24 UWB & CS-Depth
  • 34.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Future implications Use of BLE transmitters for additional low- cost tracking Simplification of systems installation at a laboratory by managing all the labs through a common broker, database and application that will be hosted remotely Custom equipment (PCB, connectors) for the moving- device to significantly reduce its size Various patterns on the moving device to be detected via the Camera Server on various device orientation-angles for optimal & robust tracking Fine-tuning of localization methods and fusion techniques by extensive testing using robots, drones and humans in various indoor environments Integrate Neural Networks in the Pattern Detection, Noise Filtering and Fusion processes Integration of additional localization methods to increase the range of this project’s applications 25
  • 35.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Future implications Use of BLE transmitters for additional low- cost tracking Simplification of systems installation at a laboratory by managing all the labs through a common broker, database and application that will be hosted remotely Custom equipment (PCB, connectors) for the moving- device to significantly reduce its size Various patterns on the moving device to be detected via the Camera Server on various device orientation-angles for optimal & robust tracking Fine-tuning of localization methods and fusion techniques by extensive testing using robots, drones and humans in various indoor environments Integrate Neural Networks in the Pattern Detection, Noise Filtering and Fusion processes Integration of additional localization methods to increase the range of this project’s applications 25
  • 36.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Future implications Use of BLE transmitters for additional low- cost tracking Simplification of systems installation at a laboratory by managing all the labs through a common broker, database and application that will be hosted remotely Custom equipment (PCB, connectors) for the moving- device to significantly reduce its size Various patterns on the moving device to be detected via the Camera Server on various device orientation-angles for optimal & robust tracking Fine-tuning of localization methods and fusion techniques by extensive testing using robots, drones and humans in various indoor environments Integrate Neural Networks in the Pattern Detection, Noise Filtering and Fusion processes Integration of additional localization methods to increase the range of this project’s applications 25
  • 37.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Future implications Use of BLE transmitters for additional low- cost tracking Simplification of systems installation at a laboratory by managing all the labs through a common broker, database and application that will be hosted remotely Custom equipment (PCB, connectors) for the moving- device to significantly reduce its size Various patterns on the moving device to be detected via the Camera Server on various device orientation-angles for optimal & robust tracking Fine-tuning of localization methods and fusion techniques by extensive testing using robots, drones and humans in various indoor environments Integrate Neural Networks in the Pattern Detection, Noise Filtering and Fusion processes Integration of additional localization methods to increase the range of this project’s applications 25
  • 38.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Future implications Use of BLE transmitters for additional low- cost tracking Simplification of systems installation at a laboratory by managing all the labs through a common broker, database and application that will be hosted remotely Custom equipment (PCB, connectors) for the moving- device to significantly reduce its size Various patterns on the moving device to be detected via the Camera Server on various device orientation-angles for optimal & robust tracking Fine-tuning of localization methods and fusion techniques by extensive testing using robots, drones and humans in various indoor environments Integrate Neural Networks in the Pattern Detection, Noise Filtering and Fusion processes Integration of additional localization methods to increase the range of this project’s applications 25
  • 39.
    Design and implementationof an integrated and modular system for 3D position calculation of a moving object using heterogeneous sensors and filters 01/04/2024 Thank you! Questions? 26