🚙 Parking Space Detection Project using Image Processing and Object Recognition

Team Name: 기둥 뒤에 공간 있어요!
Project Duration: August 20, 2024 - September 9, 2024
Team Members:

• 마석빈 (Team Lead): Model building, visualization development, data preprocessing.
• 김서희: Data analysis, YOLO data preprocessing, model training, log management.
• 박달님: Data preprocessing, model training, image augmentation.
• 이윤제: Data preprocessing (excluding YOLO), model training, log management, image augmentation.

🔍 Project Overview

This project focuses on detecting parking spaces using both traditional image processing techniques and deep learning-based object recognition algorithms. Our objective is to develop an AI model for autonomous vehicles capable of accurately detecting and classifying parking spaces, driveable spaces, pedestrians, and vehicles.

🛠️ Model Implementation Summary

Model	Backbone	Purpose	Training Details	Performance	Challenges
Mask R-CNN	ResNet-50 (ImageNet)	Object detection and instance segmentation for Parking Spaces and Driveable Spaces	- Classes: 2 (Parking Space, Driveable Space) - Epochs: 10 - Batch Size: 32 - Optimizer: Adam (lr=0.001)	Achieved 82.7% AP for parking space and driveable area detection	None
YOLOv8	N/A	Enhance detection of smaller objects like Pedestrians and Vehicles	- Classes: 2 (Person, Vehicle) - Integrated with Mask R-CNN for better performance	Achieved 74.4% AP with 4-class setup (Parking Space, Driveable Space, Person, Vehicle)	False positives (e.g., misclassifying trees or poles as people)

💻 Technical Stack

Programming Languages & Frameworks

Computer Vision & GPU

📼 Demo

👉 Watch the demo video in here!

---

📊 Dataset & References

• AI Hub Parking Space Dataset: AI Hub Data Link
• Reference 1: https://dacon.io/en/competitions/official/235672/codeshare/1795
• Reference 2: https://lyclyc52.github.io/SANeRF-HQ/