FUNDAMENTALS OF FOG AND EDGE COMPUTING

Applications and Issues

EXPLOITING FOG COMPUTING IN HEALTH MONITORING
AN ARCHITECTURE OF A HEALTH MONITORING IOT-BASED SYSTEM WITH FOG COMPUTING

• Health monitoring IoT systems have to be reliable because their results indirectly or directly
impact the medical doctor’s analysis and decisions.
• An error or a delay in the results may lead to serious consequences, such as an incorrect treatment
or a late response to the emergency, which may negatively affect human health.
• A late notification from a human fall detection monitoring IoT system to a medical doctor can lead
to a late response to a serious head injury, which quite possibly causes a death.
• In this situation, if a medical doctor was informed about the case in real time, the doctor could
provide first-aid procedures (e.g., stopping bleeding) to save the patient’s life.
• Health-monitoring IoT systems must provide high-quality data in real
time.
• The latency requirements of e-health signals vary, depending on the characteristics of
particular e-health signals.
• Maximum latency of EMG (Electromyogram) signals is less than 15.6 ms
• An architecture of the system with fog computing is shown in the below figure.
• The system includes several primary components such as
– a sensor layer
– smart gateways with a fog layer
– cloud servers with end-user terminals.
i) Device (Sensor) Layer
• A device (sensor) layer consists of sensor nodes in which each node often has three primary
components, including sensors, a micro- controller, and a wireless communication chip.

• In some applications, an SD card can be integrated into a sensor node

for storing temporary data
• The sensors (e.g., ECG, glucose, SpO2, humidity, and temperature sensors) are used for collecting
contextual data from surrounding environments and e-health data from a human body
• The contextual data such as room temperature, humidity, and statuses of the patient activities
helps to improve the quality of e-health data and a doctor’s decisions.
• For example, a heart rate of 100 beats/s of a healthy person will be normal when the person is
running, whereas this rate will be high and problematic if he/she is resting on a chair.
• Collecting contextual data does not dramatically increase the burden of
• a sensor node in termsof weight, size, complexity, and energy
consumption dramatically.
• For example, the activity significances can be extracted from a single IC chip having 3-D
accelerometer and 3-D gyroscope, while room temperature and humidity can be collected from
another IC chip.

• These chips are often small, lightweight, and energy efficiency.

• The sensors often communicate with a micro-controller via one of the wire protocols such as
Universal Asynchronous Receiver-Transmitter (UART), Serial Peripheral Interface (SPI), or Inter-
Integrated Circuit (I2C).
• The micro-controller is often a low-power chip supporting sleep modes
and waking up methods.
• The micro-controller’s frequency can vary depending on applications.
• For example, an 8 MHz micro-controller can be used for collecting high- quality data from sensors
and performing some light computation tasks (e.g., Advanced Encryption Standard – AES
algorithm) while it still fulfills requirements of latency.

• The micro-controller also communicates with a wireless chip via one of the mentioned wire
methods.
• The wireless communication chip will vary depending on the applications’ requirements.
• In general, low-power wireless protocols (e.g., BLE and 6LoWPAN) are more preferred for low data
rate applications such as fall detection or heart rate monitoring because these protocols have a
maximum bandwidth of 250 kbps.
• Wi-Fi is chosen for high-quality streaming applications (e.g., video surveillance or 24-channel EEG
monitoring) where energy consumption is not the most important criteria.
ii) Smart Gateways with Fog Computing
• Fog computing can be described as a convergent network of interconnected smart
gateways with fog services.
• Depending on applications’ requirements, a smart gateway can be
movable or fixed in a specific place.
• Each gateway type (i.e., mobile or fixed type) has its own advantages and disadvantages.
• For example, a mobile gateway provides a mobility support but it has a
limited battery capacity and hardware resource constraints.
• A fixed gateway is often built from a
powerful device supplied from a wall
socket’s power.
• The fixed gateway can perform heavy
computational tasks easily and provide
more advanced services with high-quality data while a
mobile gateway might not be able to perform the similar
tasks.
• In general, the fixed gateway is more preferred in many
healthcare applications such as remote health-monitoring
systems in the hospital and at home.
• Each smart gateway in a fog layer is an embedded device that often consists of three main
components such as
– hardware
– an operating system
– software
• Depending on specific health-monitoring applications and sensor nodes,
hardware can be various.
• For instance, a smart gateway’s wireless communication chip is compatible with the wireless
protocols (6LoWPAN, BLE, or Wi-Fi) used by sensor nodes.
• A smart gateway can be equipped with a hard drive or an SD card for storing data and installing an
operating system.
• Although a storage capacity of the hard drive or an SD card is various, it
is often not very large (e.g., less than 128 GB).
• Lightweight operating systems are often preferred in smart gateways
because it does not require powerful hardware.
• Lightweight versions of Linux kernels are used in many smart gateways.
• The operating system provides a platform for installing useful software
easily and helps to manage tasks and hardware resources more
efficiently and precisely.
• Software in a smart gateway can consist of both
– Basic programs
– Fog services.
• These programs and services are designed for serving particular
applications’ requirements such as
– Latency
– Bandwidth
– Interoperability.
• Basic programs provide fundamental features and functions of a
gateway such as
– Data transmission
– Gateway management
– Basic levels of security.
• For example, IPtable, which is a lightweight and simple software installed in Ubuntu, is used for
blocking unused communication ports of the gateway.

• MySQL or MongoDB, which is an open-source database, can be installed in the smart gateway for
flexibly, reliably, and efficiently managing databases
• Fog services can consist of
many advanced services for
enhancing the quality of
healthcare service.
• The services can help to
– reduce the burden of
sensors nodes for
extending their battery
lifetime
– saving the network
bandwidth
– reducing the burden of cloud
– informing emergent cases in real
time.
• For example, push notification for
informing emergent is an important
service of the fog
iii) Cloud Servers and End-User Terminals
• In general, there is not much difference between the cloud of a remote health-monitoring IoT
systems having a fog layer and the cloud of other IoT applications without a fog layer (e.g.,
automation, education, and entertainment).
• They all provide the fundamental features and basic services of a cloud
(e.g., data storage and data analysis).
• The burdens of the cloud in IoT systems with a fog are less than in the IoT applications without a
fog.
• For instance, ECG feature extraction algorithms and machine-learning algorithms can be
processed at a fog while the rest of the processing can be run on a cloud.
• Results from the processing are updated in both the fog’s local storage
and the cloud.
• A large amount of transmitted data can be saved and the cloud’s storage can be efficiently used.
• In general, a cloud of IoT systems with a fog is often customized for
supporting the fog services.
• For example, in the IoT systems without a fog, cloud servers do not send data back to gateways.
• In most of the cases, cloud servers simply transmit commands and
instructions to gateways, which are then forwarded to actuators.
• Cloud servers in IoT systems with a fog also transmit data to smart gateways for serving some of
the fog services such as mobility support.
• In most of the conventional health-monitoring IoT systems, Web browsers, and mobile
applications are the primary terminals of the health-monitoring IoT systems.
• These terminals are often ease-to-use, popular, and suitable for most of the devices including
smart devices (e.g., smart phone, Ipad) and computers (e.g., laptop and desktop).
• End users can access real-time data in human-readable forms (e.g., text
or graphical waveforms) via these terminals anytime and anywhere.
• In some health-monitoring IoT systems, executable programs are used together with other
terminals for accessing the monitored data.
• For example, for reducing the risks from security attacks, end users have to use a virtual private
network (VPN) and virtual platforms in order to use executable programs installed in the hospital’s
system for accessing patient’s data.
Smart Surveillance Video Stream
Processing at the Edge for Real -
Time Human Objects Tracking
What is Edge Computing?

• Edge computing refers to the practice of processing data near the

edge of the network, where the data is generated, rather than in a
centralized data-processing warehouse.

• This approach reduces latency, conserves bandwidth, and improves

response times, making it ideal for real-time applications like
surveillance.
Smart Surveillance

• Smart surveillance systems use advanced technologies such as

machine learning, artificial intelligence, and computer vision to
analyze video streams in real-time.
• These systems can detect, track, and analyze human objects,
providing valuable insights for security, crowd management, and
other applications.
Real-Time Human Objects Tracking

• Real-time human objects tracking involves detecting

and following human figures within a video stream as
they move.
• This is crucial for applications like security monitoring,
traffic management, and automated systems in smart
cities.
This diagram
demonstrates how video
data flows from cameras
through preprocessing,
human detection, and
tracking stages, and
finally to analysis, alert
generation, and storage,
with edge devices
playing a crucial role in
processing.
Key Components

[Link] Data Acquisition: Cameras and sensors capture video data.

[Link] Preprocessing: Initial processing to enhance video quality and
reduce noise.
[Link] Detection: Algorithms like Histogram of Oriented Gradients
(HOG) and Support Vector Machines (SVM) are used to detect human
figures.
[Link] Tracking: Techniques like Kernelized Correlation Filters (KCF)
track the detected objects over time.
[Link] Processing: All these steps are performed at the edge of the
network, close to the data source, to ensure real-time processing.
Video Data Acquisition
• Types of Cameras and Sensors
1. Traditional CCTV Cameras: These are the standard cameras used in many surveillance systems.
They provide continuous video feeds but often lack advanced processing capabilities.
2. IP Cameras: Internet Protocol (IP) cameras offer higher resolution and can be easily integrated
into networked systems. They allow for remote access and control.
3. Thermal Cameras: These cameras capture infrared radiation, which is useful for detecting heat
signatures, especially in low-light or nighttime conditions.
4. Smart Cameras: Equipped with built-in processors and memory, smart cameras can perform edge
processing, reducing the need to send all data to a central server.
[Link] Cameras: These capture information across different wavelengths,
such as visible light and infrared, providing more detailed and comprehensive
data.
6. LIDAR Sensors: Light Detection and Ranging (LIDAR) sensors use laser pulses to create high-
resolution maps of the environment, which can be useful in tracking movement and identifying
objects.
Data Acquisition Process

[Link] Capture: The cameras and sensors continuously capture frames of video
data.
[Link]: This step involves enhancing the video quality, such as adjusting
brightness, contrast, and reducing noise. It may also include converting the raw
data into a standard format.
[Link]: To manage bandwidth and storage, the video data is often
compressed using codecs like H.264 or H.265.
[Link]: The captured video is transmitted to storage devices or edge
processors over wired or wireless networks. IP cameras, for example, transmit
data over Ethernet or Wi-Fi.
[Link]: Video data is stored on local storage devices or network-attached
storage (NAS) systems. In edge computing, data can also be temporarily stored on
the edge devices themselves.
Data Preprocessing
• Data Preprocessing involves enhancing video quality and reducing
noise before further analysis.
• This step is crucial for ensuring that the subsequent processes like
human detection and object tracking are accurate and efficient.
Human Detection

• Human Detection involves identifying human figures within video frames. Two common
algorithms used for this purpose are:
[Link] of Oriented Gradients (HOG)
1. Description: HOG is a feature descriptor used to detect objects by analyzing the gradients in an
image.
2. Process:
1. Compute gradients in the image to detect edges and corners.
2. Create a histogram of these gradients.
3. Use this histogram as a feature vector for detecting objects.
3. Advantages: Effective for detecting humans in various poses and appearances.
[Link] Vector Machines (SVM)
1. Description: SVM is a supervised learning model used for classification.
2. Process:
1. Use feature vectors (like HOG descriptors) to train the SVM classifier.
2. The classifier learns to differentiate between human and non-human objects.
3. Advantages: Highly accurate in distinguishing between different classes.
Object Tracking

• Object Tracking involves following the detected human objects as

they move across video frames. One popular technique is:
[Link] Correlation Filters (KCF)
1. Description: KCF is a tracking algorithm that uses correlation filters.
2. Process:
• Initialize the tracker with the object's position in the first frame.
• In subsequent frames, use correlation filters to find the object's new position.
3. Advantages: Fast and efficient, suitable for real-time tracking.
Edge Processing
• Edge Devices: These are devices located close to the data sources
(e.g., cameras, sensors) that perform local processing. Examples
include smart cameras, edge servers, and IoT devices.
• Edge Gateways: Gateways connect edge devices to the network and
often provide additional processing power. They can also perform
data aggregation and filtering before sending data to the cloud or
central servers.
• Local Storage:Edge devices often have storage capabilities to
temporarily hold data. This reduces the need for constant data
transmission to central servers, conserving bandwidth and improving
efficiency.
• Processing Algorithms:Algorithms for data preprocessing, human
detection, and object tracking are executed on the edge devices. This
ensures real-time analysis and decision-making.
Benefits

• Reduced Latency: Processing data locally minimizes delays.

• Bandwidth Conservation: Less data needs to be transmitted to the
cloud.
• Enhanced Privacy: Sensitive data can be processed locally without
being sent to remote servers.
• Scalability: Edge devices can be added as needed without
overloading the network.
Applications

• Security: Real-time monitoring of public spaces, airports, and other

sensitive areas.
• Traffic Management: Analyzing pedestrian and vehicle flow in urban
areas.
• Smart Cities: Enhancing public safety and urban planning through
intelligent surveillance.
Conclusion
• Data is captured and preprocessed at the edge.
• Human figures are detected and tracked in real-time.
• Local analysis and storage occur on edge devices.
• Relevant data and alerts are transmitted to the cloud for further
analysis or long-term storage.
• Edge processing is a powerful approach that enhances the efficiency,
speed, and security of data-intensive applications. By leveraging edge
computing, smart surveillance systems can deliver real-time insights
and actions, making them indispensable for modern security and
monitoring solutions.
[Link]
atch?v=3uo1T8PYyQc