SENTINEL

Project Proposal and Project

Management
 Project Proposal and Management

Project Proposal and Management

1.1 Identification and Justification of Need 1.2 Design Brief

Residential break-ins commonly occur when homes are unoccupied, creating Sentinel is a Python desktop application designed to run on any spare Windows,
anxiety for homeowners and renters during holidays, extended work trips, or macOS or Linux laptop. It will use the laptop's built-in webcam for motion
other periods away from home. Existing home security solutions are often detection, triggering a user-selected audio alert through the laptop's speakers
expensive, with many smart alarm systems requiring ongoing subscription fees and sending a notification to the user's phone. All processing occurs locally,
or dedicated hardware. In addition, permanently installed systems are not meaning no footage or event data is ever sent to a third-party.
always suitable for renters, who may be restricted from modifying their
property. The application interface allows the user to select their camera input, choose
Many households already own compatible devices such as laptops, tablets, or and preview alert sounds, and view a live detection feed.
smartphones with built-in cameras, speakers, batteries, and Wi-Fi connectivity.
These existing features present an opportunity to develop a desktop security
application that repurposes available technology into a practical and affordable
home monitoring solution, reducing the need to purchase dedicated security
equipment.

Evaluation: The identified need is a genuine problem affecting a broad

proportion of Australian homeowners and renters. The solution:
repurposing existing hardware through software, cost-free for the user. It Evaluation:
directly addresses every required point: affordability, privacy, ease of use
and zero subscription cost.

Target Market
The target market for Sentinel is deliberately broad, as the only requirement is access to a compatible device that many Australian households already own, such
as a smartphone, tablet, or computer. By repurposing existing devices rather than requiring specialised security hardware, Sentinel provides a low-cost and
accessible home monitoring solution. This makes the software particularly suitable for renters, who may be unable to install permanent security systems, as well as
budget-conscious individuals and families seeking an affordable alternative to commercial alarm systems. The application is also designed to be easy to use,
allowing people of different ages, technical abilities, and living situations to have home security without the extra cost of security equipment or ongoing
subscription fees.

Gosford High School — Design & Technology 2026 | 2

 Areas of Investigation

Areas of Investigation Rationale for Investigation Process of Investigation

Motion detection algorithms and Core function depends on the detection method. Making
OpenCV sure the trade-offs and the accuracy, false alarm rate and
SPU usage is essential.

Existing home security software and Product needs a degree of difference from existing Research existing products: Ring Alarm, Yale Smart Living,
products solutions and must offer a unique solution to make a new TP-Link Tapo and Frigate NVR. PMI tables analysis and
product document findings and design implications.

Push Notification services Sending a real time alert to the user’s phone is a key Research and compare different services such as Ntfy,
feature. Identifying a free, privacy respecting service that Pushbullet and Gotify. MORE TESTING IS NEEDED
needs no account creation.

User needs Understanding what users want from a security app Design a Google Forms survey. Analyse results and
ensures the design is user centred. incorporate findings directly into criteria for success and
feature prioritisation.

Python desktop UI (CustomTkinter) The interface must be simple enough for a non- Research Tkinter, CustomTkinter and PyQt5. Due to use of
technological centred user to use effectively. Choosing CustomTkinter already I am familiar with it. (need to
the right UI framework affects development speed and reword.)
visual quality as well as design ease.

CPU and battery performance on The software must run smoothly on a spare, older laptop Test motion detection scripts on at least two different
older laptops without causing overheating or excessive battery drain. laptop models. Looking at the CPU usage and battery usage
over a 30 min period.

Alert and user customisation The alert sound is the primary output. Users must be able
to choose from multiple options or upload their own
effect.

Gosford High School — Design & Technology 2026 | 4

 Criteria to Evaluate Success

Category Criteria Test Method Importance

Function Detects human scale motion within 3-5 m of the Test at 1m, 3m, 5m over 5 tests in daylight and 5 at low/no light. 10/10
laptop camera Record the trigger rate.

Function Alert activates within 3 seconds of detection Measure time from motion trigger to alert sound . 10/10

Function Push notification received on phone within 10 Trigger detection and measure the time it takes for user to 10/10
seconds recieve notification on time on test device.

Function False positive rate below 5% under standard Run the system for 30 minutes with no actual motion. Count the 9/10
indoor lighting false positives.

Aesthetics Setup completable by a first time user without Ask testers to complete their setup and record their time and 8/10
major hicups. errors they faced.

Aesthetics Interface clarity and ease of use Ask users to rate the interface as well as receive feedback and 7/10
areas of improvement from them

Ergonmoics App runs stably consuming minimal CPU and Run the app continuously for a overnight test. Monitor CPU and 10/10
battery usage. battery usage using Task Manager.

Cost The app is free to download with no subscription Verify app can be installed and fully functioning on a fresh 10/10
required laptop.

Gosford High School — Design & Technology 2026 | 7

 Finance Plan

Planned Budget

Material / Technology Cost Licence/Source Notes

Python $0.00 Free - [Link] Core programing language. Cross platform compatibalitiy.

OpenCV library $0.00 Free - Apache 2.0 Motion detection, webcam capture and frame processing

CustomTkinter UI $0.00 Free - MIT licence Modern looking UI that is simple to design with Python.

Pygame $0.00 Free - LGPL licence Alert sound. Supports .mp3 and .wav files.

Ntfy push notification service $0.00 Free Phone notifications. No account required.

PyInstaller $0.00 Free Packages the app into a .exe app for easy download.

Github $0.00 Free Tracks development history. Provides a fallback if

development goes wrong.
 Time & Action Plan

Evaluation
Project Development and
Realisation
 Idea Generation

Idea #1: Standalone Physical Model Sensor

Design Concept
A wall-mounted security device comprising a motion sensor, an Arduino
microcontroller, a piezo buzzer, and LED status indicators, all housed in a
durable protective enclosure. The device is designed to operate independently
and is powered by a portable power bank, eliminating the need for it to be
plugged in and allowing it to be installed in a wide range of locations. When the
motion sensor detects movement within its detection range, the Arduino
immediately processes the signal, triggering the piezo buzzer to sound as an
audible alarm while illuminating an LED to provide a clear visual indication that
motion has been detected. The system is entirely self-contained and does not
require an internet connection, smartphone, mobile application, or additional
software to function, ensuring reliable operation even in areas without network
access. Installation is straightforward, with the enclosure mounted to a wall
using adhesive brackets, allowing for quick setup without drilling or permanent
modifications.

PMI Table

Positive Negative Implication

The device is easy to install using adhesive mounting The device relies on a power bank, which must be The sensor is an affordable, portable, and easy to use. It
brackets, requiring no drilling or permanent fixtures. It is recharged regularly to ensure its always working. It may is well suited for users who want a simple motion alarm
simple to use with no setup, internet connection, occasionally trigger false alarms due to pets, moving without relying on internet connectivity or smartphones.
smartphone, or additional software required. When objects, or other environmental factors. As the system However, its effectiveness is limited by the need for
motion is detected, it immediately activates an audible operates independently without an internet connection, regular power bank charging, the possibility of false
alarm to alert nearby users. Powered by a portable it cannot notify the owner remotely if an intrusion alarms, the lack of remote notifications, as well as the
power bank, the sensor can be easily moved around and occurs. The adhesive mounting brackets may lose their gradual weakening of adhesive mounts.
used in different locations as needed. strength over time, requiring replacement or
reattachment.

Gosford High School — Design & Technology 2026 | 2

 Idea Generation

Idea #2: Python Security App

Design Concept
A free Python-based desktop application that can run on any spare Windows,
macOS, or Linux laptop, converting it into a home security system. The
application uses the laptop's webcam to monitor its surroundings, running a
motion-detection program that looks for human movement. When motion is
detected, the software immediately plays an audio alert through the laptop's
speakers to deter potential intruders and attract the attention of anyone
nearby. At the same time, it sends a notification to the user's smartphone,
allowing them to be informed. Because the system reuses a device the user
already owns, no specialised security hardware is required, making it an
affordable and environmentally friendly alternative to other home security
systems. The application is designed to be simple to install and configure,
enabling users with minimal technical knowledge to quickly set up and begin
monitoring their home.

PMI Table

Positive Negative Implication

The application has no hardware or software cost, The laptop's lid must remain open for the webcam to The application provides an affordable and accessible
making it a completely free home security solution. It monitor the surrounding area, which limits where the home security solution by reusing a spare laptop, making
transforms a spare laptop that the user already owns device can be placed and may make it more noticeable. it a good option for users who do not want to purchase
into a functional motion detection security system, Webcam-based motion detection can also produce false dedicated security hardware. Local processing also
eliminating the need to purchase dedicated equipment. positives, with changes in lighting, shadows, pets, or enhances user privacy by keeping all footage and data
All motion detection and processing are performed other movement potentially triggering unnecessary on the device. However, the need to keep the laptop lid
locally on the device, ensuring that no footage or alerts. These false alarms may reduce the system's open limits placement options, and false positives can
personal data is transmitted to external servers. This reliability and cause users to become less responsive to reduce user trust by generating unnecessary alerts.
improves user privacy while reducing dependence on notifications over time.
cloud services or internet connectivity.

Gosford High School — Design & Technology 2026 | 2

 Idea Generation

Idea #3: Hybrid System - Physical Sensor + Software Alert

Design Concept
This design combines a dedicated motion sensor unit with a software
application running on the user's own laptop, creating a hybrid home security
system. The physical sensor provides reliable and accurate motion detection
without relying on a laptop webcam, reducing the likelihood of false positives
caused by lighting changes or camera limitations. When motion is detected, the
sensor communicates with the laptop application, which manages configurable
alerts, notification settings, and sends push notifications directly to the user's
smartphone. Because the motion sensor is housed in a separate enclosure, it
can be wall-mounted in an optimal location while the laptop remains
elsewhere, providing greater flexibility in placement.

PMI Table

Positive Negative Implication

Gosford High School — Design & Technology 2026 | 2

 Degree of Difference
Comparative Feature Analysis — Sentinel vs Existing Products
Feature / Design Factor Ring Alarm Tapo Camera Frigate NVR Sentinel

Ongoing cost $4–10/month (Cheapest) Free for local only. Paid Tapo Free (optional $50/yr Frigate+) Free
Care for full features

Data storage / privacy Cloud (Amazon servers) Local (SD card) or optional cloud 100% local 100% local

Extra hardware needed Yes - camera/doorbell hardware Yes - camera hardware Yes - dedicated PC + AI None - Reuses old hardware
accelerator (Needs to be
powerful to run AI)

Setup difficulty Simple DIY Simple DIY Complex (Docker, RTSP config) Simple install, guided UI

Works without internet No - most functions need Wi-Fi No - live view/alerts need Wi-Fi Yes for detection; notifications Yes (local alert); notification
need Wi-Fi needs Wi-Fi

Customisable alerts Limited Limited Limited Yes - choose or upload own

sound

Analysis of Key Points of Difference

Zero cost No specialised hardware Privacy by design
Unlike Ring and Tapo, which gate core Sentinel is the only option that turns a device the All motion detection and processing happens on
functionality behind a subscription, Sentinel has user already owns into a security system, rather the user's own laptop. No footage or data is ever
no paid tier at all — every feature works from day than requiring a camera, doorbell, mini-PC, or AI transmitted to a third party - matching Frigate's
one at no cost. accelerator to be purchased. privacy standard without its hardware cost or
technical barrier.

Accessible to non-technical users Personalised alerts

Frigate offers similar local-processing privacy, but its setup (Docker, RTSP The ability to preview and choose (or upload) a custom alert sound is a small
streams, self-hosting) puts it out of reach for standard people. Sentinel's but genuinely distinguishing feature - none of the three competitors
interface makes the same level of privacy achievable for a first-time user. researched offer this level of alert customisation.

Gosford High School — Design & Technology 2026 | 12

 Exploration of Existing Ideas

Product 1 — Ring
Positive Negative Implication

Wireless, modular system with real-time smartphone Core features require a paid Ring Protect subscription ($4– A wireless modular system that includes a
notifications and optional 24/7 monitoring. Simple DIY 10/month), significantly increasing long-term cost. Data is stored on
installation with no drilling for most components. Amazon cloud servers Ring has faced critisicm over footage sharing
Clean, neutral white aesthetic suitable for home with third parties without user consent. Requires stable Wi-Fi
display. Large brand ecosystem allowing expansion meaning unreliable internet makes most functions non functional
with cameras and doorbells.

Product 2 — TP-Link Tapo

Positive Negative Implication

Tapo's range is built around a subscription-free product where The subscription-free claim applies mainly to local storage. Full functionality
footage can be stored locally on a microSD card at no extra cost, such as 30-day history and notifications still requires users to pay for Tapo
with cloud storage offered as an optional extra. AI detection is also Care, so the no fees is conditional. Most models depend on a stable Wi-Fi
avalable, distinguishing people, pets and vehicles to reduce false connection for live viewing and app alerts, so a home internet outage
alerts, and entry-level models such as the C120 are priced really removes the remote-monitoring. As a camera product, Tapo also raises
well, making the brand highly accessible. Built-in sound alarms privacy concerns which may concern users who only want an intrusion alert
allow the device to deter intruders, not just record them. rather than ongoing video recording.

Product 3 — Frigate NVR

Positive Negative Implication

Frigate is a free, open-source NVR that runs on the user's Frigate's biggest problem is its complexity and hardware cost. Running AI
own hardware, keeps all recordings local, and uses AI detection on a standard CPU is expensive, so a dedicated AI accelerator
object detection to tell the difference between a person, and a mini PC server are required for smooth usage. Without a Coral
a car, a dog and a shadow. It identifies people, cars, dogs, accelerator, CPU-only detection is limited to about 2–3 cameras even on a
cats and packages, all processed locally, with no cloud mid ranged PC. Setup involves Docker, RTSP camera configuration, and self-
API calls, and no privacy concerns. The core software is hosting knowlegde that puts it out of reach for a regular person. There is
MIT-licensed, free forever, with no subscription and it also an optional paid tier, Frigate+ that costs $50/year for custom-trained
offers advanced features like license plate recognition, detection models.
face recognition.

Gosford High School — Design & Technology 2026 | 11

 Experiment and Testing

Main Function: Test #1 - Core Motion Detection Algorithm

Observations Experimentation
AIM: To test whether a basic frame-differencing algorithm using OpenCV can reliably detect
Successful human-scale motion, as this forms the core function that the rest of Sentinel depends on.
Distance Lighting Trials
Detections
METHOD: The script captures consecutive frames from the webcam, calculates the pixel
1m Daylight 5 5/5 difference between them, and applies grayscale conversion to isolate areas of change. A box is
drawn around detected changes, and any box with an area greater than 1000px² is shown as
‘motion’.
1m Low light 5 5/5
Following the test method set out in the Criteria for Success table, motion was tested at 1m,
3m, and 5m from the camera, with 5 trials in daylight and 5 trials in low/no light at each
3m Daylight 5 5/5 distance, recording the trigger rate.

3m Low light 5 5/5

5m Daylight 5 4/5

5m Low light 5 4/5

Evaluation and Need for Improvement

The algorithm successfully detected motion under all conditions at varying distances, Two limitations were identified during testing. First, the algorithm performs poorly in
meeting the success criterion. However, performance dropped in low-light trials, where low-light conditions, as it relies on pixel-value differences between frames, which
the reduced contrast between frames meant smaller or slower movements produced shrink when overall lighting is dim. Second, the method cannot distinguish between an
weaker pixel differences, sometimes falling below the detection threshold. Through actual intruder and any other moving object, such as a curtain, pet, or shadow, since it
this testing, I have found that basic frame-differencing algorithms are viable for this only detects that a change occurred, not what caused it. This means the false-positive
task, but lighting-dependent reliability needs addressing before this becomes the final rate criterion (below 5%) is unlikely to be met. Possible improvements for the next test
detection method used in the product. iteration include: adjusting the contour area threshold to filter out small/irrelevant
motion, adding a region-of-interest mask to ignore areas like windows where curtains
move, and testing other methods of detection.
Gosford High School — Design & Technology 2026 | 2
 Experiment and Testing

Main Function: Test #2 - HOG Human Detection

Observations Experimentation
AIM: Test #1 showed that frame-differencing alone cannot tell the difference between an actual
Human Human person and any other moving object (curtains, pets, shadows), which puts the false-positive rate
Successful
Distance Lighting Trials Detecion Detection really high. This test investigates the usage of OpenCV's built-in HOG (Histogram of Oriented
Detections
(In frame) (Partial) Gradients) person detector on top of motion detection and whether it could solve that problem.

1m Daylight 5 5/5 100 % 20 % Method: The script uses OpenCV's pre-trained HOG descriptor with the default people-
detecting SVM (supervised machine learning) to scan each webcam frame for human shapes. If
one or more bounding boxes are returned, the program prints "Human Detected!". This was
1m Low light 5 5/5 100 % 20 %
tested by moving in and out of frame at varying distances, angles, and levels of visibility to test
how consistently it recognised a person. Items such as a water bottle and stuffed animals were
3m Daylight 5 5/5 100 % 0% later used to see if it would recognise non-human movement as human.

3m Low light 5 5/5 80 % 0%

5m Daylight 5 4/5 80 % 0%

5m Low light 5 4/5 40 % 0%

Across repeated trials, detection was highly inconsistent. The detector only registered "Human
Detected!" when the full body was clearly visible, upright, and facing roughly toward the
camera. Partial visibility (e.g. only the upper body, or a person side-on), fast movement, or the
person being close to the edge of frame frequently produced "No Human" even though a
person was clearly present. Detection also appeared to slow down the overall frame rate
noticeably compared to Test #1's simple frame-differencing

Evaluation and Need for Improvement

Judged against the criteria for success, particularly the requirement to detect motion Two issues emerged from this test. First, HOG's requirement for a mostly full, upright, front-
within 3–5m and keep the false-positive rate below 5%, HOG detection performs facing view of a person makes it unreliable for realistic home scenarios, where an intruder is
poorly. While it tries to address the limitation identified in Test #1, it still produces a more likely to be partially obscured, side-on, or moving quickly through the frame. Second,
running HOG detection on every frame demanded more resources than simple frame-
high false-negative rate, where intruders may go undetected because they aren't fully differencing, which raises concerns about the "runs stably on older laptops with minimal CPU
or clearly in frame. For a home security application, missing a real intrusion is arguably usage" point of the criteria. Following this, test three is likely going to include another human
worse than an occasional false alarm. detection method or an improved version of the current HOG detection with improvement such
as having HOG check on the specific region of the frame where motion was detected, rather
than the whole frame. This could reduce false positives from non-human movement (like
curtains) while avoiding the reliability and performance problems of running HOG alone on
every frame.
Gosford High School — Design & Technology 2026 | 2
 Experiment and Testing

Main Function: Test #3 - Hybrid Motion + YOLO Human Detection

Observations Experimentation
AIM:
Human Human Non-
Successful
Distance Lighting Trials Detecion Detection human METHOD:
Detections
(In frame) (Partial) Detection

1m Daylight 5 5/5 100 % 20 %

1m Low light 5 5/5 100 % 20 %

3m Daylight 5 5/5 100 % 0%

3m Low light 5 5/5 80 % 0%

5m Daylight 5 4/5 80 % 0%

5m Low light 5 4/5 40 % 0%

text

Evaluation and Need for Improvement

text text

Gosford High School — Design & Technology 2026 | 2