Vehicle Detection and Parking Space Estimation using Spatial Analysis and Edge Computing for Temporary and Unorganized Lots

Managing vehicle parking in temporary and unorganized lots poses significant challenges due to the lack of fixed slot markings and irregular layouts. Existing systems rely heavily on structured environments and static infrastructure, making them unsuitable for dynamic use cases like event venues or construction sites. This paper introduces a novel AI-driven vision-based parking space estimation system that employs YOLOv8 for real-time vehicle detection, homography-based spatial mapping, and DBSCAN clustering to dynamically estimate occupancy. Unlike traditional approaches, our method requires no predefined slot information and is optimized for deployment on edge devices, enabling scalable, low-latency operation. Experimental results demonstrate high precision ($\mathbf{0. 8 9}$) and recall (0.84), validating its effectiveness across diverse real-world conditions. The proposed solution fills a key gap in current research by offering a robust, adaptive framework tailored for unstructured parking environments.