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 iterations at Unicamp.
The system has evolved through multiple research phases, leading to significant improvements in deep learning-based parking detection.
- The YOLOv11m model, optimized with TensorFlow Lite, was selected for deployment.
- Achieved the best balance between accuracy and inference speed for real-time edge computing.
- Results and a comparison between the other research phases and deployment were made and submitted to the Urban Computing Workshop (IX Workshop de Computação Urbana - CoUrb 2025) in the paper 10 Years of Deep Learning for Vehicle Detection at a Smart Parking : What has Changed?
- Conducted a benchmark study of the latest YOLO models (YOLOv8 to YOLOv11).
- Tested across multiple devices to assess performance and efficiency.
- Results were published at the arxiv paper submitted to Elsevier Internet of Things Smart Parking with Pixel-Wise ROI Selection for Vehicle Detection Using YOLOv8, YOLOv9, YOLOv10, and YOLOv11.
- Evaluated YOLO models (YOLOv3) and two-stage detectors (Mask R-CNN).
- Focused on improving accuracy and inference speed for real-time detection.
- Results of YOLOv3 were published at the technical report SmartParking A smart solution using Deep Learning.
- The system was first deployed using the SSD-based EfficientDet d2 model optimized with TensorFlow Lite.
- The project appeared in the media: Inova Campinas 2019.
- Initial research focused on CNN-based object detection for parking space identification.
- Experimented with GoogleLeNet and Xception for feature extraction.
- A presentation on the project was delivered at PAPIs.io LATAM 2018.
- Showcased early findings on smart parking and deep learning applications.
For complete hardware instructions go to 📖 Hardware Documentation
Key components:
For detailed software documentation go to 📖 Software Documentation
Key components:
- A benchmark of different deep learning models for accuracy and inference time
- Instructions to set and monitor InfluxDB