AKACHUKWU, JESSE KANAYOCHUKWU (2026) DESIGN AND IMPLEMENTATION OF AN AI-BASED VISUAL ANOMALY DETECTION SYSTEM. Other thesis, Godfrey Okoye University, Enugu Nigeria.
![[img]](https://eprints.gouni.edu.ng/style/images/fileicons/text.png) |
Text
VADS Documentation Final.pdf Download (1MB) |
Abstract
There exist several safety and security challenges arising from anomalous incidents in surveillance and industrial monitoring contexts. Contemporary strategies for visual anomaly detection within organizations in Nigeria depend significantly on traditional methods such as human led patrols, manual reviews of past footage, and visual inspections. In this work, we develop and implement a Visual Anomaly Detection System (VADS) based on computer vision and deep learning techniques for detecting two specific categories of anomalous events within video streams, namely fall incidents and abandoned objects. Fall detection is achieved using a MobileNetV2 based pose estimation pipeline that analyses human posture and movement patterns to identify fall events in real time, while abandoned object detection is achieved using an MOG2 background subtraction module that isolates static foreground objects left unattended over a defined time threshold. Model inference is exposed through a backend service for integration with edge devices, and system interaction is provided through a React based analyst dashboard that supports camera stream monitoring, alert notifications, and event logging. The system was evaluated through a structured testing process covering functional, integration, and interface testing. Ten functional test cases were executed against the core modules, namely video ingestion, fall detection, abandoned object detection, alert generation, and dashboard display, all of which produced the expected outcomes, confirming that each module performs its intended function correctly both individually and as part of the integrated pipeline. Integration testing confirmed reliable data flow between the detection modules, the alert engine, and the dashboard, with detected anomalies consistently triggering accurate, correctly timestamped alerts. Interface testing verified that the dashboard correctly displayed live camera feeds, anomaly alerts, and historical logs across the test scenarios examined. Overall, the results demonstrate that VADS is functionally reliable, modular in its design, and capable of detecting fall and abandoned object anomalies using standard camera hardware without the need for specialized sensors, making it a practical and affordable reference architecture for visual anomaly monitoring in Nigeria.
| Item Type: | Thesis (Other) |
|---|---|
| Subjects: | Q Science > Q Science (General) |
| Divisions: | Faculty of Computing And Information Technology (FACIT) |
| Depositing User: | Cynthia Ugwuoti |
| Date Deposited: | 24 Jul 2026 14:49 |
| Last Modified: | 24 Jul 2026 14:49 |
| URI: | http://eprints.gouni.edu.ng/id/eprint/6011 |
Actions (login required)
 |
View Item |
