An IoT-Based Smart Parking Management System with Automatic Number Plate Recognition and Environmental Hazard Monitoring

Abstract

Parking is a growing problem in urbanizing Nepali
cities, where many parking lots still rely on manual occupancy
recording, vehicle logging, and cash transactions. Although
past IoT applications have looked into parking systems, Nepali
ANPR, reservation management, and automatic parking payments
have been investigated separately in earlier research,
there are very few cases where all these aspects are combined
together along with environmental hazard detection into a single,
low-cost physical model.This paper presents a connected IoTbased
smart parking system using ESP32/Arduino sensors and
actuators, a React.js dashboard backed by Firebase RTDB, and
a tiered Nepali ANPR pipeline. Plate localization and character
detection are performed using YOLOv8s, a fine-tuned 34-class
CNN, and geometrical reconstruction to assemble individually
detected characters into a full plate string. The object detection
framework had 95.44% mAP@0.50 (97.92% & 93.09% for Recall
and precision, respectively) after held-out tests on a dataset
containing 867 images. Overall, the ANPR Pipeline had a
whole-plate detection accuracy of 53.73% on raw 67 entry/exit
pictures without cropping (36/67; 95% CI: 41.9-65.1%), using
fuzzy whole-plate character matching with an 80% threshold.
In view of the small number of end-to-end test data, this finding
should be seen more as an initial performance estimate than as
a statistically reliable indication of deployment-based accuracy
for ANPR. The occupancy detection component performed at
97.14% accuracy from 70 test samples manually inspected, while
the reservation, transaction log maintenance, and hazard detection
capabilities were also demonstrated on the real-world system
without quantitative evaluation. These outcomes demonstrate the
feasibility of combining localized ANPR technology with low-cost,
IoT-enabled parking infrastructure in Nepal, although large-scale
implementation should await further end-to-end evaluation and
field trials.