{"id":{"repo_id":"uwtsd","oai_identifier":"oai:repository.uwtsd.ac.uk:4241"},"canonical_url":"https://search.dev.ndltd.org/etd/uwtsd/oai:repository.uwtsd.ac.uk:4241","repository":{"repo_id":"uwtsd","name":"University of Wales Trinity Saint David","base_url":"https://repository.uwtsd.ac.uk/cgi/oai2"},"display":{"title":"An Ai-Powered 3-D Facial Recognition Framework for Crime Intelligence using Reconstructive and Predictive Techniques","abstract":"Police use of facial recognition tends to be divided between stand-alone biometric tools and discrete dashboards for analysing crime, with very little integration between who appears in images and how patterns of crime develop. At the same time, mainstream 2-D recognition models are brittle under CCTV conditions, where pose, blur, compression and occlusion are common. This dissertation explores the feasibility of a 3-D aware facial analytics pipeline to enhance its robustness on unconstrained imagery, and how its outcomes can be fused with the results of crime trend analytics in a research-only prototype. The proposed framework comprises YOLOv8s person/weapon detector, landmark, FLAME-based 3-D ReconstructionNet trained on AFLW2000-3D, and RecognitionNet trained on 530 FaceScrub actor/actress public individuals. RecognitionNet generates 512-D embeddings using an LDA-reduced 256-D version and tests using closed set identification, verification ROC, robustness and explainability tools (Grad-CAM, region occlusion). Nevertheless, a comparison suite is provided where these embeddings are compared to Buffalo InsightFace features using k-NN classification, perclass diagnostics and UMAP visualisation. Crime-trend analytics built from anonymised UK open police data in Facebook/Meta Prophet, with the use of a full-stack investigator dashboard feeding from REST endpoints. Governance components simulate the automated DPIA, encryption and audit logging as well as bias monitoring over male/female FaceScrub partitions. Experiments demonstrate that the 3-D reconstruction stage has low vertex and landmark errors, as well as being stable to moderate pose and occlusion. RecognitionNet has moderate closed-set accuracy but good verification performance (AUC > 0.83) and is much better than the Buffalo baseline in terms of the k-NN embedding quality on the project’s domain. Prophet is able to capture the clear structure of season crime and support hotspot-style visualisation within the frontend. Bias monitoring reveals small but nonnegligible differences in demographic performance, which argues in favour of continuous fairness measurement. The work shows that the technical feasibility of an integrated 3-D facial recognition and crime analytics stack based on public, anonymised data is possible, but also highlights the various barriers to be overcome in terms of legal, ethical, and data quality considerations, before the creation of any operational solution.","abstract_html":"Police use of facial recognition tends to be divided between stand-alone biometric tools and discrete dashboards for analysing crime, with very little integration between who appears in images and how patterns of crime develop. At the same time, mainstream 2-D recognition models are brittle under CCTV conditions, where pose, blur, compression and occlusion are common. This dissertation explores the feasibility of a 3-D aware facial analytics pipeline to enhance its robustness on unconstrained imagery, and how its outcomes can be fused with the results of crime trend analytics in a research-only prototype. The proposed framework comprises YOLOv8s person/weapon detector, landmark, FLAME-based 3-D ReconstructionNet trained on AFLW2000-3D, and RecognitionNet trained on 530 FaceScrub actor/actress public individuals. RecognitionNet generates 512-D embeddings using an LDA-reduced 256-D version and tests using closed set identification, verification ROC, robustness and explainability tools (Grad-CAM, region occlusion). Nevertheless, a comparison suite is provided where these embeddings are compared to Buffalo InsightFace features using k-NN classification, perclass diagnostics and UMAP visualisation. Crime-trend analytics built from anonymised UK open police data in Facebook/Meta Prophet, with the use of a full-stack investigator dashboard feeding from REST endpoints. Governance components simulate the automated DPIA, encryption and audit logging as well as bias monitoring over male/female FaceScrub partitions. Experiments demonstrate that the 3-D reconstruction stage has low vertex and landmark errors, as well as being stable to moderate pose and occlusion. RecognitionNet has moderate closed-set accuracy but good verification performance (AUC > 0.83) and is much better than the Buffalo baseline in terms of the k-NN embedding quality on the project’s domain. Prophet is able to capture the clear structure of season crime and support hotspot-style visualisation within the frontend. Bias monitoring reveals small but nonnegligible differences in demographic performance, which argues in favour of continuous fairness measurement. The work shows that the technical feasibility of an integrated 3-D facial recognition and crime analytics stack based on public, anonymised data is possible, but also highlights the various barriers to be overcome in terms of legal, ethical, and data quality considerations, before the creation of any operational solution.","abstract_has_math":false,"creators":["Paraschiv, Cristian Constantin"],"institution":"University of Wales Trinity Saint David","degree_name":"msc","degree_level":"masters","degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2026,"date_issued":"2026-03","date_published":"2026-03","updated_at":"2026-07-24T05:53:11Z","subjects":["HV Patholeg gymdeithasol. Lles cymdeithasol a chyhoeddus","QA76 Meddalwedd Cyfrifiadurol"],"languages":[],"rights":[],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier.grantnumber","label":"Dc Identifier Grantnumber","values":["UWTSD"],"render_values":[{"text":"UWTSD","href":null,"code":true}]}]},"links":{"outbound_url":"https://doi.org/10.82227/repository.uwtsd.ac.uk.00004241","outbound_label":"DOI","outbound_source":"dc:identifier.doi"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.sponsor","label":"Sponsor","values":["University of Wales Trinity Saint David"]},{"key":"dc:creator","label":"Author","values":["Paraschiv, Cristian Constantin"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2026-03-25"]},{"key":"dc:date.issued","label":"Date","values":["2026-03"]},{"key":"dc:publisher.commercial","label":"Dc Publisher Commercial","values":["University of Wales Trinity Saint David"]},{"key":"dc:publisher.department","label":"Dc Publisher Department","values":["Traethodau Meistr","Institute of Inner City Learning"]},{"key":"dc:publisher.institution","label":"Dc Publisher Institution","values":["University of Wales Trinity Saint David"]},{"key":"dc:relation.isreferencedby","label":"Dc Relation Isreferencedby","values":["https://repository.uwtsd.ac.uk/id/eprint/4241/"]},{"key":"dc:type","label":"Dc Type","values":["Gosodiad"]},{"key":"dc:type.qualificationlevel","label":"Dc Type Qualificationlevel","values":["masters"]},{"key":"dc:type.qualificationname","label":"Dc Type Qualificationname","values":["msc"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["HV Patholeg gymdeithasol. Lles cymdeithasol a chyhoeddus","QA76 Meddalwedd Cyfrifiadurol"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.doi","label":"DOI","values":["10.82227/repository.uwtsd.ac.uk.00004241"]},{"key":"dc:identifier.grantnumber","label":"Dc Identifier Grantnumber","values":["UWTSD"]},{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://repository.uwtsd.ac.uk/id/eprint/4241/1/Paraschiv_CC_MSc_Thesis.pdf"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Police use of facial recognition tends to be divided between stand-alone biometric tools and discrete dashboards for analysing crime, with very little integration between who appears in images and how patterns of crime develop. At the same time, mainstream 2-D recognition models are brittle under CCTV conditions, where pose, blur, compression and occlusion are common. This dissertation explores the feasibility of a 3-D aware facial analytics pipeline to enhance its robustness on unconstrained imagery, and how its outcomes can be fused with the results of crime trend analytics in a research-only prototype. The proposed framework comprises YOLOv8s person/weapon detector, landmark, FLAME-based 3-D ReconstructionNet trained on AFLW2000-3D, and RecognitionNet trained on 530 FaceScrub actor/actress public individuals. RecognitionNet generates 512-D embeddings using an LDA-reduced 256-D version and tests using closed set identification, verification ROC, robustness and explainability tools (Grad-CAM, region occlusion). Nevertheless, a comparison suite is provided where these embeddings are compared to Buffalo InsightFace features using k-NN classification, perclass diagnostics and UMAP visualisation. Crime-trend analytics built from anonymised UK open police data in Facebook/Meta Prophet, with the use of a full-stack investigator dashboard feeding from REST endpoints. Governance components simulate the automated DPIA, encryption and audit logging as well as bias monitoring over male/female FaceScrub partitions. Experiments demonstrate that the 3-D reconstruction stage has low vertex and landmark errors, as well as being stable to moderate pose and occlusion. RecognitionNet has moderate closed-set accuracy but good verification performance (AUC > 0.83) and is much better than the Buffalo baseline in terms of the k-NN embedding quality on the project’s domain. Prophet is able to capture the clear structure of season crime and support hotspot-style visualisation within the frontend. Bias monitoring reveals small but nonnegligible differences in demographic performance, which argues in favour of continuous fairness measurement. The work shows that the technical feasibility of an integrated 3-D facial recognition and crime analytics stack based on public, anonymised data is possible, but also highlights the various barriers to be overcome in terms of legal, ethical, and data quality considerations, before the creation of any operational solution."]},{"key":"dc:format","label":"Dc Format","values":["text"]},{"key":"dc:title","label":"Title","values":["An Ai-Powered 3-D Facial Recognition Framework for Crime Intelligence using Reconstructive and Predictive Techniques"]}]}],"canonical_facts":{"dc:contributor.sponsor":["University of Wales Trinity Saint David"],"dc:creator":["Paraschiv, Cristian Constantin"],"dc:date":["2026-03-25"],"dc:date.issued":["2026-03"],"dc:description.abstract":["Police use of facial recognition tends to be divided between stand-alone biometric tools and discrete dashboards for analysing crime, with very little integration between who appears in images and how patterns of crime develop. At the same time, mainstream 2-D recognition models are brittle under CCTV conditions, where pose, blur, compression and occlusion are common. This dissertation explores the feasibility of a 3-D aware facial analytics pipeline to enhance its robustness on unconstrained imagery, and how its outcomes can be fused with the results of crime trend analytics in a research-only prototype. The proposed framework comprises YOLOv8s person/weapon detector, landmark, FLAME-based 3-D ReconstructionNet trained on AFLW2000-3D, and RecognitionNet trained on 530 FaceScrub actor/actress public individuals. RecognitionNet generates 512-D embeddings using an LDA-reduced 256-D version and tests using closed set identification, verification ROC, robustness and explainability tools (Grad-CAM, region occlusion). Nevertheless, a comparison suite is provided where these embeddings are compared to Buffalo InsightFace features using k-NN classification, perclass diagnostics and UMAP visualisation. Crime-trend analytics built from anonymised UK open police data in Facebook/Meta Prophet, with the use of a full-stack investigator dashboard feeding from REST endpoints. Governance components simulate the automated DPIA, encryption and audit logging as well as bias monitoring over male/female FaceScrub partitions. Experiments demonstrate that the 3-D reconstruction stage has low vertex and landmark errors, as well as being stable to moderate pose and occlusion. RecognitionNet has moderate closed-set accuracy but good verification performance (AUC > 0.83) and is much better than the Buffalo baseline in terms of the k-NN embedding quality on the project’s domain. Prophet is able to capture the clear structure of season crime and support hotspot-style visualisation within the frontend. Bias monitoring reveals small but nonnegligible differences in demographic performance, which argues in favour of continuous fairness measurement. The work shows that the technical feasibility of an integrated 3-D facial recognition and crime analytics stack based on public, anonymised data is possible, but also highlights the various barriers to be overcome in terms of legal, ethical, and data quality considerations, before the creation of any operational solution."],"dc:format":["text"],"dc:identifier.doi":["10.82227/repository.uwtsd.ac.uk.00004241"],"dc:identifier.grantnumber":["UWTSD"],"dc:identifier.uri":["https://repository.uwtsd.ac.uk/id/eprint/4241/1/Paraschiv_CC_MSc_Thesis.pdf"],"dc:publisher.commercial":["University of Wales Trinity Saint David"],"dc:publisher.department":["Traethodau Meistr","Institute of Inner City Learning"],"dc:publisher.institution":["University of Wales Trinity Saint David"],"dc:relation.isreferencedby":["https://repository.uwtsd.ac.uk/id/eprint/4241/"],"dc:subject":["HV Patholeg gymdeithasol. Lles cymdeithasol a chyhoeddus","QA76 Meddalwedd Cyfrifiadurol"],"dc:title":["An Ai-Powered 3-D Facial Recognition Framework for Crime Intelligence using Reconstructive and Predictive Techniques"],"dc:type":["Gosodiad"],"dc:type.qualificationlevel":["masters"],"dc:type.qualificationname":["msc"]},"updated_at":"2026-07-24T05:53:11Z"}