Traffic Image Realtime prediction!

Let’s start by looking into the data,

Data Selection

This week, I chose Singapore LTA realtime data from CCTV cameras available online,

https://beta.data.gov.sg/collections/354/view

I started by exploring the data, and looking for potential applications. Then I found that We can measure the congestion of each location using object detection since the data are associated with spatial location of each camera.

Problem definition

• Build an interface that shows the road congestion in Singpaore using the available Data.gov.sg open dataset from CCTV roads.
• Visualize the data on a live platform.

Model selection

I have done a quick research on the best models that can be used to detect objects with high accuracy, and eventually, I chose YoloV5.

Tools

• Visualization: Deck.gl
• Python 3.11
• Pytorch
• VsCode
• Firebase.

Training/Testing

First, I used Pytorch Yolov8 pretrained model to build my model on cars using Udacity self driving cars dataset availble here.

• How to replicate:
  • Create an account on roboflow,
  • Goto roboflow Universe
  • Search for the car detecting and how many model,
  • Click download the model
  • Choose the model YoloV8.
  • Copy/paste the code snippet.
  • Now, you are ready to download the trained model.

Building the pipeline

Deployment on GCP

Live preview