Experiment: 0

Problem Statement
Name: Adarsh Bhargava

Roll-no: 0187AS231005

Version: 1.0

Department: CSE-AIDS

Object Detection System

In an increasingly digital and automated world, object detection has emerged as a core
component of many cutting-edge technologies including autonomous vehicles, surveillance
systems, smart cities, retail analytics, and robotics. The ability to accurately identify and
locate objects within images or video streams is essential for machines to understand and
interact with their environment intelligently.

The challenge lies in enabling machines to process visual data and recognize multiple object
categories in real-time, under varying conditions such as occlusion, lighting, scale, and
motion. Traditional image processing methods lack the robustness and adaptability
required for such complex tasks. Hence, deep learning-based object detection algorithms
like YOLO (You Only Look Once), SSD (Single Shot Detector), and Faster R-CNN have been
developed to meet these needs.

This project aims to build an Object Detection System that uses modern convolutional
neural networks (CNNs) to detect and classify multiple objects within images or video
input. The system will be trained on annotated datasets such as COCO or Pascal VOC and
will be capable of drawing bounding boxes around detected objects with labels and
confidence scores.

The core objective is to create an efficient and accurate detection system that can operate in
real-time with minimal false positives and false negatives. It should be scalable to different
domains, such as identifying people in surveillance footage, detecting damaged goods in
warehouses, or spotting traffic signs for autonomous vehicles.

Challenges include optimizing model performance for real-time inference, handling dataset
imbalance, addressing edge case variability, and ensuring robustness across diverse
scenarios. Techniques such as data augmentation, transfer learning, and post-processing
(e.g., non-max suppression) will be explored to improve the model's performance.

This project not only contributes to practical applications in computer vision but also
deepens understanding in neural networks, model evaluation, and deployment in real

world systems.
Experiment :1

🎭 Actors:
User – Individual interacting with the object detection system (e.g., using a mobile or web
application).

Admin – Person managing and maintaining the system backend, datasets, and model
updates.

📌 Use Cases for User:

Upload Image/Video: Allow users to input image or video files for analysis.

Live Camera Feed Detection: Detect objects in real-time through a connected camera.

View Detection Results: Display detected objects with bounding boxes and labels.

Download Detection Report: Allow users to save output as annotated images or reports.

Select Detection Categories: Enable users to focus detection on specific object types (e.g.,
vehicles, people, animals).

🔧 Use Cases for Admin:

Manage Object Categories: Add or remove classes that the detection model should
recognize.

Model Update & Retraining: Retrain model using updated datasets or improved algorithms.

Monitor System Performance: Track detection accuracy, latency, and error logs.

Dataset Management: Add new annotated data to improve model generalization.

Use Case Diagram :-
Software Requirement Specification (SRS)
Project Title: Object Detection System

Date: 12-06-2025

Version: 1.0

By: Adarsh Bhargava

Roll-No: 0187AS231005

Table of Contents
• 1. Introduction
• 1.1 Product Scope
• 1.2 Product Value
• 1.3 Intended Audience
• 1.4 Intended Use
• 1.5 General Description
• 2. Functional Requirements
• 3. External Interface Requirements
• 4. Non-Functional Requirements
• 5. Definitions and Acronyms

1. Introduction

1.1 Product Scope

The Object Detection System is a software application designed to detect and classify
multiple objects within images or video feeds using deep learning models such as YOLO or
SSD.

1.2 Product Value

- Enables automated surveillance, safety systems, and inventory checks.
- Reduces human effort and error in object tracking and classification.
- Can be deployed in real-time applications such as autonomous vehicles and security
systems.

1.3 Intended Audience

- Developers and researchers working with computer vision or AI.
- Companies requiring real-time object tracking solutions.
- Educational institutions for project-based learning.
1.4 Intended Use
- Detect and identify multiple objects in an image/video.
- Track object movements frame-by-frame in video.
- Label and save results for further use or analytics.

1.5 General Description

The system utilizes pre-trained convolutional neural networks for detecting and identifying
various object classes. Users can upload images or connect a live video feed. The results are
visualized with bounding boxes and labels.

2. Functional Requirements

2.1 User Use Cases

• Upload Media: Allow users to upload image/video for object detection.
• Live Detection: Use webcam or camera to detect objects in real time.
• View Results: Display detected objects with labels and confidence scores.
• Download Output: Save detection results as images or JSON files.
• Feedback: Submit feedback for incorrect classifications.

2.2 Admin Use Cases

• Model Management: Update, switch, or train new detection models.
• Data Analytics: View detection logs and statistics.
• System Monitoring: Monitor model performance and server health.

3. External Interface Requirements

3.1 User Interface

Simple, responsive UI with image/video upload and output visualization panels.

3.2 Hardware Interface

- Webcam or IP camera for real-time detection.
- GPU-accelerated hardware (optional but recommended).

3.3 Software Interfaces

- TensorFlow, PyTorch, OpenCV, Flask/Streamlit for deployment.
- Integration with external detection APIs (optional).

3.4 Communication Interfaces

- RESTful APIs for remote detection services.
- WebSocket for live streaming updates.

3.5 Data Interfaces

- Input: JPEG, PNG, MP4 formats.
- Output: JSON (object classes, coordinates), labeled images.
4. Non-Functional Requirements

4.1 Performance
Detect and label objects in less than 1 second per frame on GPU.

4.2 Reliability
99.5% system uptime, with fallback model support.

4.3 Scalability
Able to process batch image/video uploads.

4.4 Security
- Secure image upload and access permissions.
- Encrypted data transmission.

4.5 Usability
Clean, intuitive interface with tooltips and help section.

4.6 Maintainability
Modular, well-documented codebase with unit tests.

4.7 Compliance
Follows open-source licensing guidelines and GDPR for data usage.

5. Definitions and Acronyms

• Object Detection: Locating and classifying objects within visual data.
• YOLO: You Only Look Once – a fast object detection model.
• SSD: Single Shot Detector – another efficient object detection architecture.
• CNN: Convolutional Neural Network.
• JSON: JavaScript Object Notation.
• REST API: API using HTTP methods to interact with web services.
• FPS: Frames Per Second, used in video processing.
Experiment :03
Class Diagram
 Experiment :03
Class Diagram

Class Diagram Description:

The Object Detection System consists of the following primary classes:

1. Camera
- Attributes: resolution, frameRate, deviceID
- Methods: captureFrame()

2. ImageProcessor
- Attributes: preprocessingParams
- Methods: preprocessImage(image)

3. Detector
- Attributes: modelType, confidenceThreshold
- Methods: detectObjects(image)

4. BoundingBox
- Attributes: x, y, width, height, label, confidence
- Methods: drawBox(), getCoordinates()

5. ResultManager
- Attributes: detectionResults
- Methods: saveResults(), displayResults()

Relationships:
- Camera captures input and sends it to ImageProcessor.
- ImageProcessor preprocesses the input and forwards it to Detector.
- Detector outputs BoundingBox instances for detected objects.
- ResultManager handles and visualizes the final detection output.
 Experiment-4
Installation of Miniconda

(1). Download Process

(1.1).
(2). Miniconda3 Installing

(3). Conda List Fetching Process

AD 403 Name: Adarsh Bhargava Roll-no:0187AS231005

Experiment No 05

Title : Creation of Conda Environment

1. Creation of Conda Environment

2. Verification of Conda Environment

1
AD 403 Name: Adarsh Bhargava Roll-no:0187AS231005

3. Activate Cando Environment

4. Deactivate Cando Environment

5. Output _ Conda Info –envs