Smart Parking Assistant : Chatbot for Real-time Parking Space Finder

Abstract

This thesis explores an innovative approach to address urban parking challenges through the development of a Smart Parking Assistant. The system combines advanced computer vision techniques with a conversational AI interface to facilitate efficient parking space detection and user interaction.

The core of the system utilizes a YOLOv5l object detection model, trained on both real and synthetic parking lot data. This approach resulted in high accuracy, with the model achieving 97.9% precision and 96% recall in identifying parking spaces. The integration of synthetic data significantly enhanced the model's ability to generalize across various parking scenarios.

A Rasa-based chatbot complements the vision system, providing a user-friendly interface for parking information retrieval. The chatbot demonstrated 92% intent recognition accuracy with an average response time under 2 seconds, contributing to an intuitive user experience.

Key challenges encountered during development included the model's initial limitations with unannotated parking spaces and performance variability in diverse real-world footage. These issues were addressed in some extent through fine-tuning with synthetic data and the application of transfer learning techniques.

The research highlights the potential of AI-driven solutions in urban parking management, offering a scalable and potentially cost-effective alternative to traditional sensor-based systems. However, it also underscores the need for further advancements in real-time processing, environmental adaptability, and integration with broader smart city infrastructures.

Future work will focus on enhancing real-time capabilities, expanding feature sets, and conducting extensive field testing. This project contributes to the growing field of smart city technologies and opens avenues for further research in AI-driven urban mobility solutions.