BUILD AN AI OBJECT DETECTION USING RASPBERRY PI

Department of Electronics & Communication Engineering

Submitted by

REGISTER
NAME
NUMBER

University College Of Engineering,Villupuram

A Constitution College Of Anna University,Chennai

Guided by

Mr. S. SURYA

TRAINER
 ABSTRACT
This project introduces an intelligent object detection
system meticulously crafted for real-time monitoring, automation, and
edge computing applications, all powered by the versatile Raspberry
Pi platform. By harnessing the synergistic power of computer vision
and deep learning methodologies, the system strategically employs a
Raspberry Pi Camera module in conjunction with a streamlined
TensorFlow Lite model. This integration enables the accurate
detection and identification of a diverse range of objects within the
camera's field of view. Operating locally on the Raspberry Pi's
processing capabilities, the system offers a compact, remarkably cost-
effective, and highly energy-efficient solution, ideally suited for a
multitude of applications including intelligent surveillance systems,
sophisticated home automation setups, and a wide array of Internet of
Things (IoT) deployments. By significantly reducing reliance on
cloud-based processing, the system inherently bolsters user privacy,
minimizes latency in object recognition, and maintains operational
capabilities even in environments characterized by limited or
intermittent network bandwidth. This research underscores the
potential of edge AI in delivering practical and efficient computer
vision solutions.
TABLE OF CONTENT:

S.NO CONTENT

1. ABSTRACT

2. INTRODUCTION

3. PROBLEM STATEMENT

4. PROPOSED SYSTEM

5. COMPONENTS REQUIRED

6. SOFTWARE

7. BLOCK DIAGRAM

8. PIN DIAGRAM & OUTPUT

9. WORKING

10. CONCLUSION
 CHAPTER-1

INTRODUCTION
The field of Artificial Intelligence (AI) has ushered in a
transformative era, fundamentally altering the way machines perceive,
interpret, and interact with their surrounding environments. Within the
vast landscape of AI domains, object detection emerges as a critical
capability, empowering machines with the ability to not only "see" but
also to identify and localize specific entities within digital images or
video streams. This capacity forms the bedrock for a plethora of
advanced applications, ranging from autonomous vehicles navigating
complex road scenarios to sophisticated security systems capable of
identifying potential threats.
Traditional object detection systems often rely on
computationally intensive algorithms and powerful hardware, making
their deployment in resource-constrained environments challenging.
However, the advent of lightweight deep learning models and the
proliferation of affordable single-board computers like the Raspberry
Pi have democratized access to advanced AI capabilities. This project
capitalizes on these advancements, exploring the feasibility and
effectiveness of implementing a real-time object detection system on
the Raspberry Pi. By leveraging the efficiency of TensorFlow Lite, a
streamlined version of Google's popular machine learning framework,
the system aims to achieve a balance between detection accuracy and
computational feasibility on the Raspberry Pi's hardware.
 CHAPTER-2

PROBLEM STATEMENT:

Existing object detection solutions often present

limitations in terms of cost, computational resource requirements, and
reliance on constant internet connectivity. Many sophisticated object
detection systems necessitate powerful and expensive hardware,
making them inaccessible for numerous small-scale projects and
individual users. Furthermore, cloud-based solutions, while offering
significant computational power, introduce concerns regarding data
privacy, latency issues due to network dependence, and recurring
operational costs.
In scenarios demanding real-time analysis and
immediate responses, such as autonomous surveillance or robotic
navigation in dynamic environments, the latency associated with
transmitting data to and from the cloud can be a significant bottleneck.
Moreover, in remote or bandwidth-constrained locations, the
feasibility of relying on cloud-based processing is severely limited.
This project addresses these challenges by
investigating the development of a localized, cost-effective object
detection system that operates directly on the edge, utilizing the
Raspberry Pi as its core processing unit. The primary problem is to
design and implement an AI-powered object detection system that can
achieve a reasonable level of accuracy and speed while running
efficiently on the Raspberry Pi's limited computational resources,
 minimizing the need for cloud connectivity and thereby addressing
concerns related to cost, latency, and privacy.
CHAPTER-3

PROPOSED SYSTEM:

The proposed system is an AI-driven object

detection framework meticulously designed for implementation on the
Raspberry Pi platform. At its core, the system integrates a Raspberry
Pi Camera module for capturing real-time visual data and leverages a
pre-trained, lightweight deep learning model, specifically optimized
for mobile and embedded devices through TensorFlow Lite. This
model is responsible for analyzing the incoming video frames and
identifying objects of interest within them.
The system operates on the principle of edge
computing, where the object detection processing occurs directly on
the Raspberry Pi itself, eliminating the need for continuous data
transmission to a remote server. This on-device processing enhances
privacy by keeping sensitive data local, reduces latency by avoiding
network communication delays, and ensures functionality even in the
absence of internet connectivity.
The workflow of the proposed system involves
the continuous acquisition of video frames from the Raspberry Pi
Camera. Each captured frame undergoes necessary preprocessing
steps to ensure compatibility with the input requirements of the
TensorFlow Lite model. The preprocessed frame is then fed into the
 model, which performs the object detection task, identifying and
localizing objects by drawing bounding boxes around them and
assigning corresponding class labels. The resulting annotated video
feed can be displayed on a connected monitor or further processed for
specific applications, such as triggering alerts upon the detection of
certain objects or logging the occurrences of detected entities. The
system is designed to be modular and extensible, allowing for
potential integration with other IoT devices and platforms for
enhanced functionality and automation.
CHAPTER-4

COMPONENTS REQUIRED:

❖ Raspberry Pi (Model 4B or later recommended): The central

processing unit of the system.
❖ Processor: Broadcom BCM2711, Quad core Cortex-A72 (ARM
v8) 64-bit SoC @ 1.5GHz
❖ RAM: Available in 2GB, 4GB, or 8GB variants (4GB or 8GB
recommended for smoother AI model execution)
❖ Connectivity: 2.4 GHz and 5.0 GHz IEEE 802.11ac wireless,
Bluetooth 5.0, BLE, Gigabit Ethernet, 2 × USB 3.0 ports, 2 × USB
2.0 ports
❖ Video Output: 2 × micro-HDMI ports supporting up to 4K@60fps
❖ Power: 5V DC via USB-C connector (minimum 3A recommended
for stable operation with peripherals)
❖ Operating System: Raspberry Pi OS (or other compatible Linux
 distributions)
❖ Raspberry Pi Camera Module : The sensor for capturing video
data.
❖ Sensor: 8-megapixel Sony IMX219 sensor
❖ Resolution: 3280 x 2464 pixels (still images), 1080p at 30fps,
720p at 60fps, and 640x480p at 90fps (video)
❖ Lens: Fixed focus
❖ Field of View: 62.2 x 48.8 degrees
❖ Sensor: 12.3-megapixel Sony IMX477 sensor
❖ Resolution: 4056 x 3040 pixels (still images)
❖ Lens Mount: C-mount and CS-mount (requires compatible lenses,
sold separately)
❖ Flexibility: Offers greater control over focus, aperture, and field of
view with interchangeable lenses.
❖ MicroSD Card: For storing the operating system, software
libraries, and project files.
❖ Specification: Class 10 or higher for faster read/write speeds.
❖ Power Supply: A stable power source for the Raspberry Pi and
connected peripherals.
❖ Specification: 5V DC, with sufficient current rating (e.g., 3A for
Raspberry Pi 4B).
❖ Display : A monitor or screen for viewing the real-time video feed
and output.
❖ Connectivity:HDMI cable compatible with the Raspberry Pi's
micro-HDMI port (or adapter if needed).
 ❖ USB Keyboard and Mouse: For interacting with the Raspberry
Pi's operating system.
❖ Wi-Fi or Ethernet Connection : For software installation, remote
access, and potential integration with other systems.
CHAPTER-5

SOFTWARE:

➢ Raspberry Pi OS
➢ Purpose: Recommended OS with built-in drivers and a suitable
environment for development.
➢ Installation:

1. Download Raspberry Pi Imager.

2. Flash Raspberry Pi OS onto a microSD card.
3. Insert card, boot the Raspberry Pi, and complete initial setup.

➢ Python 3
➢ Purpose: Main programming language for the application.
➢ Installation:

1. Check version: python3 --version

2. If missing, install with:

➢ TensorFlow Lite

1. Purpose: Lightweight ML framework to run pre-trained models

efficiently.
2. Installation:
➢ OpenCV (cv2)

1. Purpose: For real-time image processing and drawing detection

boxes.
2. Installation:

➢ NumPy

1. Purpose: Used for array manipulation and numerical operations.

2. Installation:

➢ Pre-trained TensorFlow Lite Model

1. Purpose: A deep learning model (e.g., MobileNet SSD) in

.tflite format with label map.
2. Download: From TensorFlow Model Zoo or online
repositories. Ensure compatibility with TensorFlow Lite.
 Applications include:

✓ Smart surveillance (detecting intrusions, tracking movement).

✓ Home automation (e.g., activating lights or alarms).
✓ Educational and research purposes in AI and embedded systems.
✓ Robotics (object detection and navigation assistance).
CHAPTER-6

COMPONENTS REQUIRED:

COMPONENT DESCRIPTION/PURPOSE
Raspberry Pi (3B+, 4, or 5) Main processing unit for running object
detection algorithms.
microSD Card (16GB+, Class 10) Stores OS, Python code, models, and
libraries. High-speed cards improve
performance.
Camera Module / USB Webcam Captures real-time video input for
detection.
Power Supply (5V 3A) Provides stable power to the Raspberry Pi.
Use official power supply for best results.
Monitor, HDMI Cable, Keyboard, Mouse Required for initial setup; optional if
using headless mode via SSH or VNC.
Heat Sink / Cooling Fan Keeps the Raspberry Pi cool during long
processing tasks.
Pre-trained TensorFlow Lite Model Lightweight .tflite model like
MobileNet SSD for object detection.
Internet Connectivity (Wi-Fi/Ethernet) Needed for downloading packages,
models, or enabling cloud integration
USB Hub Useful if multiple USB devices are
connected to the Raspberry Pi.
Raspberry Pi Enclosure / Case Protects components and assists with
heat dissipation.
 CHAPTER-7

SOFTWARE SETUP

INSTALLATION OF RASPBIAN OPERATING SYSTEM ON

RASPBERRY Pi

Installing Raspbian on the Raspberry Pi is really clear. Raspbian will be

downloaded and writing the disc image to a micro SD card, at that point
booting the Raspberry Pi to that microSD card. For this undertaking, one
needs a microSD card (with no less than 8 GB), a PC with a space for it,
and, obviously, a Raspberry Pi and fundamental peripherals (a mouse,
console, screen, and power source). This isn't the main strategy for
introducing Raspbian (more on that in a minute), yet it's a valuable
method to learn on the grounds that it can likewise be utilized to
introduce such a significant number of other working projects on the
Raspberry Pi. When one realizes how to compose a circle picture to a
microSD card, we open up a great deal of alternatives for Raspberry
Piprojects.
Step1: Download the Raspbian

Turn on the PC and download the Raspbian disc image. One can locate
the most recent variant of Raspbian on the Raspberry Pi Foundation's site
here. It will take some time, particularly in the event when one intends to
utilize the conventional download alternative as opposed to the other
download sources. It can take a stretch of half hour or more to download.
Choose Raspbian Stretch with Desktop if you want to have access to the
Raspbian GUI; in other words, if you want to log in and be able to access
a desktop, icons, etc. like you would with Windows or MacOS. Choose
Raspbian Stretch Lite if you only need to boot to the command line. For
simpler Raspberry Pi projects, this is often a good choice since the Lite
version uses less power and fewer resources.
Step2: Unzip the file
The Raspbian disc image is compressed, so it should be unzipped. The file
uses the ZIP64 format, so depending on how current built-in utilities are, one
needs to use certain programs to unzip it. Linux users will use the
appropriately named Unzip Configure Pi Camera: Enable the camera
interface via raspi-config.

Step 3: Use Etcher

One has to pop the microSD card into our computer and write the disc image
to it. The process of actually writing the image will be slightly different
across these programs, but it’s pretty self- explanatory no matter what is
being used. Each of these programs will have us select the destination and
the disc image (the unzipped Raspbian file). Choose, double-check, and then
button to write.
Step 4: Put the microSD card in your Pi and boot up
When the disc image has been kept in touch with the microSD card, it is
prepared to go. Put that microSD into your Raspberry Pi, plug in the
peripherals and power source. The present release to Raspbian will boot
straightforwardly to the desktop. Our default credentials are username pi and
password raspberry.

Step 5:
➢ Connect the Raspberry Pi to your network using a network cable, connect the
keyboard and mouse and plug in the HDMI cable between the Pi and the
Desktop.

Step 6:
After the Pi finishes booting up then the Raspbian desktop should appear.
Open the Raspberry Menu, go to Preferences, then open Raspberry Pi
Configuration
 1.
2.

STEP 7:
Check Wait for Network. This means the Pi will wait to start the
desktop, and XLink, until we have a network connected
STEP 8:
Enable SSH and VNC, this will enable us to access the Raspberry
Pi remotely without needing a TV/Keyboard later on.

1. Install TensorFlow Lite: Install the appropriate TensorFlow Lite

runtime and dependencies.
2. Install Python Libraries: Use pip to install OpenCV, NumPy, and other
required packages.
3. Run Object Detection Script:
o Capture frames from the Pi Camera.
o Use TensorFlow Lite model to detect objects.
o Display results with bounding boxes and object labels using
OpenCV.
 CHAPTER-8
BLOCK DIAGRAM

VGA DISPLAY
IMAGES

KEYBOARD

RASPBERRY PI MODEL B
MOUSE

SDCARD
CAMERA
INTERFACE

POWER
SUPPLY
 CHAPTER-9

PIN DIAGRAM & OUTPUT

➢ GPIO Pin Usage:

▪ GPIO pins can be configured to trigger external devices
such as buzzers, LEDs, or relays when specific objects are
detected.
▪ For example, a GPIO pin can activate an alarm when a
person is detected in a restricted area.
➢ Output:
▪ Real-time video feed with object labels and bounding boxes.
▪ Sample output includes detection of people, bottles,
chairs, etc., depending on the model.
▪ Optionally, the system can log detections or send alerts via
Wi-Fi to a remote server or IoT dashboard.
 CHAPTER-10

WORKING

o The system begins by initializing the camera and AI model.

o It continuously captures frames from the live video stream.
o Each frame is resized and preprocessed to meet the model’s
input specifications.
o The preprocessed frame is passed through the MobileNet SSD model.
o The model outputs bounding boxes and class labels for detected
objects.

o The results are displayed on screen with real-time annotations.

o Additional features may include:
o Alert notifications on detecting specific objects.
o Integration with cloud dashboards (e.g., Blynk, Firebase).
o Data logging for later analysis.
 CHAPTER-11
CONCLUSION

This project demonstrates how AI and edge computing can be effectively

combined to create a real-time object detection system using Raspberry Pi. The

proposed solution is compact, cost-efficient, and suitable for a wide range of

applications from personal to industrial use. By leveraging TensorFlow Lite, the

system ensures smooth performance even on limited hardware resources. It

promotes privacy by eliminating the need for cloud dependency and provides a

foundation for scalable AI-powered IoT systems. Future improvements could

include model customization, enhanced detection accuracy, and integration with

smart home ecosystems.

 REFERENCECS

[1] George E.Sakr ,Maria Mokbel,Ali Hadi and Ahmad Darwich ,COMPARING
DEEP LEARNING AND SUPPORT VECTOR MACHINES FOR AUTONOMOUS
WASTE SORTING in 2016 IEEE International Multidisciplinary Conference on
Electrical Technology (IMCET) ,vol 26,no.3,pp.657-853,March 2016

[2] Tiagarajah ,V.Janahiraman1 and Mohammad Shahrul Subuhan ,TRAFFIC LIGHT

USING TENSORFLOW OBJECT DETECTION FRAMEWORK in 2017 IEEE 9th
International Conference on System Engineering and Technology (ICSET),vol
27,no.4,pp.567-598,March 2017.

[3] Abdelhakim Zeroual ,P.F Rocha and M.P Silva,SSD MultiBox Object Detection
,published in the Proceedings of the International Conference on Networking and
Advanced Systems (ICNAS) IEEE,vol 25,no.2,pp.543-538,March 2018.

[4] Wei Liu,Dragomir Anguelov,Dumirtu Erhan,Christian Szegedy,”SSD:Single Shot

MultiBox Detector “ in arxiv Cornell University,2019.

[5] De Charette, R. and Nashashibi, F., “Traffic light recognition using image
processing compared to learning processes,” In 2009 IEEE/RSJ International
Conference on Intelligent Robots and Systems, vol 15,no.6,pp.345-214,October 2009.

[6] Liu, W., Anguelov, D., Erhan, D., Szegedy, C., Reed, S., Fu, C.Y. and Berg, A.C.,
“Ssd: Single shot multibox detector,” in ArXiv,20 Dec 2014.

[8] W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C.-Y. Fu, and A. C. Berg.
Ssd: “Single shot multibox detector”,.In European Conference on Computer Vision.