View Proposal

Proposer
Mohammad Ammad-Uddin
Title
Design, Implementation and Evaluation of AI-Based Visual Detection and Localization Using Computer Vision
Goal
To investigate, design, implement and evaluate an AI-based computer-vision system for detecting, classifying and, where appropriate, localizing objects, abnormalities or regions of interest within images or video. The student will identify and select a suitable application domain and develop an appropriate intelligent vision solution.
Description
Deep-learning and computer-vision techniques are widely used to identify and localize objects, patterns, defects and other regions of interest in images and video. These techniques have applications in healthcare, agriculture, manufacturing, environmental monitoring, transport, safety and intelligent mobile systems. This project will investigate the application of AI-based computer vision to a selected visual detection and localization problem. The student will first review potential application areas and select an appropriate problem based on practical relevance, availability and quality of data, suitability for AI-based analysis, computational requirements, ethical considerations and feasibility within the project timeframe. The selected problem will be analysed to determine appropriate input data, target classes and system requirements. The student will investigate and select suitable computer-vision and deep-learning techniques. Possible approaches may include image classification, object detection, semantic or instance segmentation, transfer learning or combinations of these techniques. The choice of method should be justified through literature review and initial experimentation rather than prescribed in advance. The technical work may include dataset preparation, preprocessing, annotation or label conversion, data augmentation, model training, parameter selection, comparison of alternative approaches and performance optimisation. A prototype should be developed that processes images, recorded video or live camera input and produces suitable visual or textual output. Depending on the selected application, the system may identify a class, localize one or more targets with bounding boxes, generate segmentation masks, detect abnormalities or report that no relevant target has been detected. Where appropriate, the final prototype may be integrated into a mobile, web-based or desktop application. Expected Project Work • Investigate and compare candidate application areas for AI-based visual analysis. • Select and justify a feasible problem, dataset and target classes. • Prepare a reproducible data pipeline, including preprocessing and any required annotation or label conversion. • Investigate and select appropriate AI/computer-vision models or architectures. • Train, tune and evaluate the selected solution and compare it with a suitable baseline or alternative. • Integrate the final model into a usable prototype and critically analyse real-world limitations. Example Applications (Illustrative Only) • Plant disease or leaf-damage detection and localization. • Road damage, pothole or infrastructure-defect detection. • Waste detection, sorting or recyclable-object recognition. • Fruit or food quality assessment using image-based features. • Industrial surface-defect or product-quality inspection. • Personal protective equipment or workplace-safety object detection. • Wildlife, pest or environmental-object monitoring. • Skin-lesion or visible-abnormality analysis as a non-diagnostic research prototype using public de-identified data and an appropriately constrained scope.
Resources
Background
Url
Difficulty Level
Easy
Ethical Approval
None
Number Of Students
3
Supervisor
Mohammad Ammad-Uddin
Keywords
artificial intelligence, deep learning, computer vision, object detection, image classification, image segmentation, visual localization, transfer learning, real-time ai
Degrees
Bachelor of Science in Computational Sciences and Software Engineering
Back to List