DEVELOPMENT OF A COMPUTER VISION-BASED SYSTEM FOR MONITORING AND DETECTING UNAUTHORISED PARKING OF TRICYCLES AND MINIBUSES IN ENUGU CITY

OKENWA, CHUKWUKA (2026) DEVELOPMENT OF A COMPUTER VISION-BASED SYSTEM FOR MONITORING AND DETECTING UNAUTHORISED PARKING OF TRICYCLES AND MINIBUSES IN ENUGU CITY. Other thesis, Godfrey Okoye University, Enugu.

Text Okenwa Chukwuka updated.pdf Download (4MB)

Abstract

In this paper, a fully automatic vision-based detection system for detection of unauthorized parking violation by Keke NAPEP tricycles and commercial minibuses in no-parking zones in Enugu metropolis, Nigeria has been proposed and evaluated. This includes designing, developing and evaluating the performance of a combination of custom-trained YOLO11s (You Only Look Once, small variant) object detection model with Region of Interest (ROI) enforcement model based on polygon, delivered through PyQt6 desktop application. Unlike the traditional rule based/sensor-based parking monitoring systems, the present system was developed by training with custom annotated data set consisting of 175 images of Keke and 115 images of minibus, manually labeled using CVAT and curated using Roboflow, making it adaptable to the visual features of the environment in Enugu city. The YOLO11s model has been fine-tuned using transfer learning on pre-trained COCO weights over 50 epochs, batch size 16 and input resolution 640x640 pixels on Google Colab GPU machine. In violation detection, a check has been made for centroid of the bounding box of the detected vehicles whether it falls inside the polygon of the illegal zone using OpenCV polygon intersection method for a number of consecutive frames. Performance evaluation of the trained model was done based on the object detection metrics; Precision, Recall, F1-Score, Average Precision (AP) and mean Average Precision (mAP) at Intersection Over Union (IoU) of 0.5 and 0.5:0.95 ([email protected] and [email protected]:0.95). For the validation data set, the trained model scored a Precision of 0.758, Recall of 0.832, [email protected] of 0.838 and [email protected]:0.95 of 0.607, above the threshold of 0.80 [email protected] as required by the project. For the class level results, the Keke class had an Average Precision of 0.946 and the bus class had an Average Precision of 0.716. The highest F1-Score of 0.78 was realized at confidence score of 0.539 and maximum combined recall of 0.98 at confidence score of near zero. These results show that the YOLO11s model has met a high and operationally robust standard of detection accuracy, especially for the Keke class. From these results, it is evident that it is technically feasible to deploy lightweight and real time computer vision systems for automated traffic law enforcement even in cities like Enugu.

Item Type:	Thesis (Other)
Subjects:	Q Science > Q Science (General)
Divisions:	Faculty of Arts > Faculty of Law > Faculty of Management and Social Sciences > Faculty of Education > Faculty of Natural and Applied Sciences
Depositing User:	MICHAEL MADUBUKO
Date Deposited:	27 Jul 2026 10:19
Last Modified:	27 Jul 2026 10:19
URI:	http://eprints.gouni.edu.ng/id/eprint/6025

Actions (login required)

View Item