This is a semester-long coursework for the third year of the Bachelor of Software Engineering program at Yoobee Colleges in Auckland, focusing on advanced topics in software development and project management.
This repository contains code for a vehicle tracking system based on the YOLOv8 object detection algorithm.The interface is powered by Streamlit. The system is designed to work in challenging scenarios, such as low light conditions, occlusions, and fast motion.
- Feature 1: Object detection task.
- Feature 2: Multiple detection models.
yolov8n,yolov8s,yolov8m,yolov8l,yolov8x,DriveWatch - Feature 3: Multiple input formats.
Image,Video,Webcam - Feature 4: Multiple Object Tracking and Counting.
You can use This link to try an online version on Streamlit.
# for Windows
# create
python -m venv myvenv
# activate
myvenv\Sripts\activate
git clone https://github.com/irinagetman1973/Vehicle-tracking-in-challenging-scenarios-using-YOLOv8.git
cd Vehicle-tracking-in-challenging-scenarios-using-YOLOv8
# Streamlit dependencies
pip install streamlit
# YOLOv8 dependencies
pip install -e '.[dev]'
Create a directory named weights and create a subdirectory named detection and save the downloaded YOLOv8 object detection weights inside this directory. The weight files can be downloaded from the table below.
| Model | size (pixels) |
mAPval 50-95 |
Speed CPU ONNX (ms) |
Speed A100 TensorRT (ms) |
params (M) |
FLOPs (B) |
|---|---|---|---|---|---|---|
| YOLOv8n | 640 | 37.3 | 80.4 | 0.99 | 3.2 | 8.7 |
| YOLOv8s | 640 | 44.9 | 128.4 | 1.20 | 11.2 | 28.6 |
| YOLOv8m | 640 | 50.2 | 234.7 | 1.83 | 25.9 | 78.9 |
| YOLOv8l | 640 | 52.9 | 375.2 | 2.39 | 43.7 | 165.2 |
| YOLOv8x | 640 | 53.9 | 479.1 | 3.53 | 68.2 | 257.8 |
My custom-trained YOLOv8 model that I named DriveWatch using the following approach:
model = YOLO('yolov8n.yaml').load('yolov8n.pt') . The model was trained on a dataset specifically curated for bright lighting conditions. The resulting model is fast and lightweight. Notably, during visual evaluation within the Streamlit app, the model exhibited better performance compared to the largest model available in similar bright lighting scenarios. A dataset of bright lighting images for testing models can be downloaded here.
For faster inference applied on videos and livestream using webcam, use lighter models. YOLOv8n (nano) works just fine.
streamlit run app.py
Then will start the Streamlit server and open your web browser to the default Streamlit page automatically.
