Real-Time Object Detection and Live Stream Analysis Using Jetson Nano

🎯 MINI PROJECT

👥 TEAM MEMBERS

• Ganesh Patidar (20214061)
• Hardik Kumar Singh (20214249)
• Divyanshu (20214317)
• Harsh Dave (20214534)

---

📌 CONTENTS

• Problem Statement
• Introduction
• Motivation
• Applications
• Proposed Work
• Experimental Setup
• Result Analysis
• Challenges
• Future Work
• References

---

📢 Problem Statement

• Livestream Camera Integration with Jetson Nano Hardware
• Object Detection on Images, Videos, and Livestream Feeds

📖 Introduction

This project implements a real-time object detection system using Jetson Nano, leveraging deep learning algorithms for accurate and efficient object classification. It enhances surveillance, security, and operational efficiency in various applications.

💡 Motivation

The inspiration for this project stems from the critical need to improve security measures in public transport systems. By leveraging real-time CCTV feeds, we aim to provide an automated surveillance system that ensures passenger safety, particularly for vulnerable groups. Our goal is to enable authorities to detect potential security threats proactively.

🚀 Applications

• Surveillance and Security Systems
• Traffic Management
• Retail Analytics
• Industrial Automation
• Smart Cities
• Environmental Monitoring

🔍 Proposed Work

• Jetson Nano Setup
• Live Stream Implementation
• Data Collection & Model Training
• Evaluation of Object Detection Models
• Performance Analysis of Different Models

🛠 Experimental Setup

1️⃣ Setting Up Jetson Nano

• Flashed the NVIDIA OS using Balena Etcher.
• Installed JetPack SDK 4.4.0 for development.
• Booted Jetson Nano and configured the environment.

2️⃣ Live Streaming Implementation

• Utilized OpenCV with CUDA for optimized real-time video processing.
• Enabled efficient video capture and frame-by-frame object detection.

3️⃣ Data Collection & Model Training

• Collected data using simple_image_download.
• Labeled images using labelImg.
• Trained a YOLOv7 model using Google Colab for improved computational performance.

4️⃣ Evaluation of Object Detection Models

• Compared TensorFlow Model Zoo models:
  • SSD ResNet50 640x640
  • CenterNet ResNet101 FPNv1 512x512
• Evaluated based on mean Average Precision (mAP) and inference time.

5️⃣ Performance Metrics

• Precision = TP / (TP + FP)
• Recall = TP / (TP + FN)
• mAP = Average of AP across all classes

📊 Result Analysis

✅ Accuracy Comparison

Model	mAP (Accuracy)
CenterNet ResNet-101	Low
SSD ResNet-50	Moderate
YOLOv7 (Custom)	High

⚡ Inference Time Trade-offs

• Fastest: CenterNet ResNet-101 (Low accuracy, high speed)
• Balanced: SSD ResNet-50 (Moderate speed & accuracy)
• Most Accurate: YOLOv7 (High accuracy, slower inference)

📷 Example Results

🛑 Challenges

• Proxy Configuration Issues
• Package Installation Errors
• SSL Wrong Version Number
• Python Version Conflicts
• Extended Training Time
• Jetson Nano Compatibility Issues
• Unexpected Shutdowns During Execution

🔮 Future Work

• Performance Optimization
• Cloud Integration
• Real-time Alerts & Notifications
• Enhanced User Interface
• IoT Device Integration

📚 References

1. Abadi, M. et al. TensorFlow Model Zoo
2. Liu, W., Anguelov, D., et al. SSD: Single Shot Multibox Detector, ECCV (2016)
3. Redmon, J., et al. YOLO: Unified, Real-Time Object Detection, IEEE TPAMI (2016)
4. Wang, J., et al. YOLOv7: Trainable Bag of Freebies, IEEE TPAMI (2021)
5. PyTorch for Jetson

---

🚀 Thank you! We appreciate your time in reviewing our project! 🎯