automotive-tracking

This code performs object detection and tracking on automotive video data.

Prerequisites

To be able to perform object detection and tracking, one must first prepare the environment and the data.

Setting up the environment

Recommended environment can be installed on Unix systems via:

# install Miniconda
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh -b -u
rm -rf Miniconda3-latest-Linux-x86_64.sh
conda list # Verify the installaton

# create separate environment and install necessary packages:
conda create -n automotive python=3.11.7 numpy
conda activate automotive
pip install ultralytics opencv-python cap_from_youtube moviepy jupyter matplotlib pandas

Obtaining the data

If just playing with YOLO model, it is enough to find a YouTube url or a local video/image file of interest and use it with /src/detect_and_track_YOLO.py or examples/detect_and_track_YOLO.ipynb.

To get the large automotive datasets with object annotations etc., one can use the download scripts, provided in their respective subdirectories in datasets/.

How to run

Simple YOLO

The object detection and tracking with YOLO can be run from the terminal:

python src/detect_and_track_YOLO.py videos/video_filename.mp4 --arg1

or by using the notebook: examples/detect_and_track_YOLO.ipynb

QUBO

This has not been implemented yet.