🚦 Traffic Management System Using IoT and Machine Learning

The Traffic Management System Using IoT and Machine Learning project aims to enhance urban traffic management by detecting real-time vehicle flow and analyzing the data to reduce congestion. Utilizing a Jetson Nano with a YOLO object detection model, the system detects vehicles such as cars, bikes, and autorickshaws. The vehicle data is transmitted to the cloud using the MQTT protocol, interfacing with AWS IoT Core. The long-term goal is to leverage this data to build a predictive model for traffic patterns and integrate it into a digital twin using VSim software.

---

📑 Project Overview

This project focuses on improving traffic flow by integrating IoT devices and machine learning algorithms. A Jetson Nano device running the YOLO object detection model identifies different types of vehicles in real-time at a 4-way intersection. Data is sent to the cloud using AWS IoT Core, where it is stored and analyzed for traffic optimization. Future work includes determining the format of data for VSim import and defining data handling workflows in AWS.

---

🛠️ Technologies Used

• Hardware: Jetson Nano, OAK-D Lite Camera
• Machine Learning: YOLO Object Detection, DeepSort
• Cloud: AWS IoT Core ( AWS Free Tier)
• Software: VSim, Python, Ubuntu

---

🚧 Project Roadmap

🖥️ Jetson Nano and YOLO Implementation

• Acquire Jetson Nano hardware
  Status: Completed ✅
• Train YOLO model on custom dataset
  Status: Completed ✅
• Integrate DeepSort into YOLO
  Status: Completed ✅
• Check compatibility of module dependencies with Jetson Nano
  Status: Completed ✅
  Note: Downgraded to python 3.6.9 on Jetson nano and built pytorch , openCV ,numpy etc from the source. Jetson nano using CUDA v10.2 , Python v3.6.9 ,Pytorch v1.10.1 and torchvision v0.11.1 Made several changes to the deep_sort and ultaralytics (YOLO V8) modules to make it compatible with these versions
• Run YOLO on Jetson Nano for real-time detection (using a stock video)
  Status: Completed ✅
  Note: But the execution is slow that we need to research on ways to optimize the execution. Already disabled the GUI interface and increased performance , but need to research on more optimizations
• Extend the model to run detections on a 4-way junction (using a stock video)
  Status: Open 🟢
• Optimize the detection in Jetson Nano
  Status: Open 🟢
• Integrate OAK-D Lite Camera on Jetson Nano
  Status: Open 🟢
• Perform real-time vehicle detection using the live camera feed
  Status: Open 🟢

☁️ Cloud Integration

• Sign up for AWS free tier and obtain credits
  Status: Completed ✅
  Note: Applied for several programmes and completed surveys for free credit .Didn't got any reply till now.

• Design cloud architecture in AWS
  Status: Completed ✅

• Register Jetson Nano in AWS IoT Core and establish a connection
  Status: Completed ✅

• Send detection results from Jetson Nano to AWS IoT Core
  Status: Open 🟢

• Define data format and transmission frequency to AWS IoT Core
  Status: Open 🟢

• Decide post-receipt processing of data in AWS IoT Core
  Status: Blocked 🔴

📁 Repository Structure

- /deep_sort       # Original deepsort code forked from repository and did modifications according to out dependencies
- /main.py     # python script for object detection