http://hdl.handle.net/20.500.12283/344 2026-04-05T16:43:58Z 2026-04-05T16:43:58Z Akol, Joseph http://hdl.handle.net/20.500.12283/4679 2026-02-06T07:24:48Z 2025-01-01T00:00:00Z

Gait recognition as a biometric signal, investigating skeletal trajectory extraction and classification approaches Akol, Joseph Gait recognition, a non-intrusive biometric method effective at a distance, serves as a practical alternative to traditional identification techniques like fingerprints or facial recognition, particularly in forensic and surveillance contexts. This study aims to design and compare 2D and 3D models for gait recognition with the general objective of analyzing their comparative performance. Specific objectives include: (1) identifying a robust gait dataset recorded across various environmental conditions and camera angles, (2) extracting 2D and 3D skeletal keypoints using OpenPose and MediaPipe, respectively, (3) developing and evaluating the accuracy of 2D and 3D gait recognition models, and (4) analyzing their performance under varying environmental conditions. The research addresses the effectiveness of dataset utilization, comparative model accuracy, and key performance influencers. Results show the 2D model excels in occlusion handling, achieving a 0% false positive rate (FPR) for unknown instances, while the 3D model better generalizes to unseen sequences at higher thresholds, with a 74.55% FPR. However, the 2D model struggles with single-frame performance (80% FPR), and the 3D model exhibits significant bias and poor occlusion handling (100% FPR). Recommendations include enhancing dataset diversity, incorporating temporal features via LSTM layers, and improving 3D keypoint robustness through occlusion augmentation. This study provides a replicable framework, offering insights into model trade-offs and advancing gait recognition for resource constrained environments, forensic applications, and academic research. Dissertation

2025-01-01T00:00:00Z Kobusinge, Irene http://hdl.handle.net/20.500.12283/4669 2026-01-06T12:24:43Z 2025-01-01T00:00:00Z

Development of magnetite zeolite nanocomposite for sustainable crude oil spill remediation Kobusinge, Irene Oil spills pose a significant threat to both marine and freshwater ecosystems, affecting biodiversity, public health, and socio-economic stability. Uganda, poised to commence commercial oil production by 2026 through the Tilenga and Kingfisher projects with an estimated output of 6.5 million barrels per day, faces heightened risks of environmental contamination—particularly in ecologically sensitive areas such as the Albertine Graben. Traditional oil spill remediation methods often prove inefficient, recovering only a small fraction of spilled oil, thereby necessitating the development of more effective and sustainable alternatives. This research investigated the effectiveness of magnetite-zeolite nanocomposites as an environmentally friendly, reusable adsorbent for crude oil spill remediation in Uganda. The nanocomposite was synthesized and characterized using several techniques, alongside adsorption isotherm studies. These analyses provided detailed insights into the material’s surface morphology, structural integrity, chemical composition, and functional group interactions. Batch adsorption experiments were conducted under varying operational conditions such as contact time, initial oil concentration, and adsorbent dosage (Eskandari et al., 2018) to evaluate the nanocomposite’s performance in terms of adsorption capacity, removal efficiency. The Response Surface Methodology (RSM) was employed to statistically optimize the process parameters using Central Composite Design (CCD) in Design Expert software and graphs drawn with origin pro software. The results demonstrated that the magnetite-zeolite nanocomposite exhibited high adsorption efficiency, magnetic recoverability, validating its potential for large-scale application in oil spill response strategies with 93% removal efficiency after Optimisation. This research contributes to the advancement of sustainable oil spill remediation technologies and aligns with Sustainable Development Goal 14 by offering an innovative and context-specific solution to Uganda’s oil pollution challenges. Dissertation

2025-01-01T00:00:00Z Akello, Winnie http://hdl.handle.net/20.500.12283/4668 2026-01-06T12:16:56Z 2025-01-01T00:00:00Z

Investigation of pectin-based adhesive from sweet potato residues for enhanced binding of nonwoven fabrics. Akello, Winnie The growing demand for sustainable and eco-friendly adhesives in the textile industry has highlighted the need to explore natural alternatives to synthetic, petroleum-based adhesives. This project investigated the effectiveness of pectin-based adhesive from sweet potato residues for binding nonwoven textile materials. Sweet potato peel, an abundant agricultural byproduct in Uganda is under-utilized and yet can be used as a low-cost source of pectin. The study focuses on three main objectives: optimizing the pectin extraction process, characterizing the chemical properties of the extracted pectin and evaluating its adhesive performance in binding nonwoven fibers. The use of Response Surface Methodology to optimize the extraction parameters mainly extraction time and extraction temperatures using Composed Composite design experiment consisting of 32 runs was employed to determine pectin yield as the response with hydrochloric acid as the extraction solvent. The pH of the extraction solvent was kept constant at 1.5. This was followed by modifying the adhesion of the extracted pectin using calcium chloride as a crosslinking agent and glycerol as plasticizer. Both modified and unmodified extracted pectin sample was analyzed on Fourier Transform Infrared Spectroscopy (for functional group and degree of esterification) which are very crucial for adhesive properties. The adhesive’s binding strength, thermal stability and water resistance tests were carried out to assess the suitability for textile applications. These properties were conducted on nonwoven cotton fabrics as hygienic nonwoven fabrics. Results from this study demonstrated the potential of sweet potato derived pectin as a renewable, environmentally friendly adhesive contributing to sustainable development in the textile industry Dissertation

2025-01-01T00:00:00Z Ntanda, Nassar http://hdl.handle.net/20.500.12283/4667 2026-01-06T12:06:17Z 2025-01-01T00:00:00Z

Diagnosing foot and mouth disease and lumpy skin disease in cattle using computer vision Ntanda, Nassar Cattle farming is a major source of income for many people in the Sub-Harran region, and the livestock sector played a vital role in supporting the region’s economy and food security [2]. However, the prevalence of infectious cattle diseases, particularly Lumpy Skin Disease (LSD) and Foot and Mouth Disease (FMD), posed serious threats to food security, farmer livelihoods, and animal productivity. Addressing the impacts of these diseases required early and accurate diagnosis [1]. This study presents a computer vision-based diagnostic model that utilizes deep learning techniques to detect and classify FMD and LSD in cattle. The proposed system captured the visible signs of infection using a Convolutional Neural Network (CNN) trained on a dataset of annotated cattle images [3]. The model achieved high levels of accuracy and robustness under diverse environmental conditions and was capable of distinguishing between healthy cattle and those infected with FMD or LSD. To improve accessibility, the trained model was integrated into a mobile application, enabling rural farmers and veterinary practitioners to perform real-time diagnosis without the need for specialized equipment. The study also addressed challenges related to image data collection in the Sub-Harran region, such as variations in cattle breeds, lighting conditions, and disease presentation. Data augmentation and transfer learning techniques were employed to improve the model’s generalization. The model’s performance will be evaluated using standard metrics, including accuracy, precision, recall, and F1-score, with results indicating strong potential for practical deployment [4]. Overall, this study will offer a scalable solution for improving cattle disease management through the application of artificial intelligence, with implications for enhancing disease surveillance, reducing economic losses, and improving animal health in resource-constrained settings. Dissertation

2025-01-01T00:00:00Z