This repository contains the complete implementation of a real-time smart parking monitoring system using edge computing and deep learning, representing a decade of research at Unicamp.
The system evolved through multiple research phases, culminating in real-world deployments optimized for accuracy, inference speed, and low-power edge devices.
We value your opinion!
If you have any complaints, compliments, or suggestions about the IC2 Smart Parking Project, please share them with us using the form below:
🌎 Português:
A sua opinião é muito importante!
Caso tenha reclamações, elogios ou sugestões sobre o Projeto IC2 Smart Parking, compartilhe conosco no formulário abaixo:
👉 Preencha o Formulário de Feedback
- Overview
- Architecture
- Project Timeline
- Demonstration
- Hardware Setup
- Software Implementation
- Features
- Collaborators
- How to Cite
- License
This system uses deep learning-based object detection to identify free and occupied parking spots in real-time. The design supports deployment on low-power devices such as Raspberry Pi, leveraging TensorFlow Lite optimizations for on-device inference.
It has undergone four major research phases since 2015, progressively improving detection accuracy, inference speed, and robustness under real-world conditions.
- Deployed YOLOv11m optimized with TensorFlow Lite.
- Achieved best trade-off between accuracy and inference speed for edge computing.
- Results published in:
- 10 Years of Deep Learning for Vehicle Detection at a Smart Parking: What has Changed? (CoUrb 2025) and received honorable mention as one of the best papers.
- Benchmark study on YOLOv8–YOLOv11.
- Multi-device performance evaluation.
- Results published at the arxiv paper submitted to Elsevier Internet of Things:
- Evaluation of YOLOv3 and Mask R-CNN.
- Focused on balancing accuracy and speed for real-time use.
- Results published in:
- Deployed SSD-based EfficientDet d2 with TensorFlow Lite.
- Featured in the media:
- Explored GoogleLeNet and Xception for CNN-based parking detection.
- Presented at PAPIs.io LATAM 2018.
Full details: 📖 Hardware Documentation
Key components:
Full details: 📖 Software Documentation
Key modules:
- Real-time detection of parking spot occupancy.
- Edge-optimized inference using TensorFlow Lite.
- Multiple model support (YOLO, EfficientDet, Mask R-CNN).
- Historical data logging with InfluxDB.
- Modular hardware and software documentation.
Present:
- Professor Juliana Freitag Borin
- Dr. Luis Fernando Gomez Gonzalez
- Gustavo P. C. P. da Luz
- Gabriel Massuyoshi Sato
- Tiago Godoi Bannwart
- Smart Campus Unicamp
Past:
@article{da2024smart,
title={Smart Parking with Pixel-Wise ROI Selection for Vehicle Detection Using YOLOv8, YOLOv9, YOLOv10, and YOLOv11},
author={da Luz, Gustavo PCP and Sato, Gabriel Massuyoshi and Gonzalez, Luis Fernando Gomez and Borin, Juliana Freitag},
journal={arXiv preprint arXiv:2412.01983},
year={2024},
doi= {10.48550/arXiv.2412.01983}
}
@inproceedings{da202510,
author = {Gustavo da Luz and Gabriel Sato and Tiago Bannwart and Luis Gonzalez and Juliana Borin},
title = {10 Years of Deep Learning for Vehicle Detection at a Smart Parking : What has Changed?},
booktitle = {Anais do IX Workshop de Computação Urbana},
location = {Natal/RN},
year = {2025},
pages = {127--140},
publisher = {SBC},
address = {Porto Alegre, RS, Brasil},
doi = {10.5753/courb.2025.8869},
url = {https://sol.sbc.org.br/index.php/courb/article/view/35256}
}
This project is licensed under the MIT License.