{"id":{"repo_id":"bournemouth","oai_identifier":"oai:eprints.bournemouth.ac.uk:41016"},"canonical_url":"https://search.dev.ndltd.org/etd/bournemouth/oai:eprints.bournemouth.ac.uk:41016","repository":{"repo_id":"bournemouth","name":"University of Bournemouth","base_url":"http://eprints.bournemouth.ac.uk/cgi/oai2"},"display":{"title":"Empower dynamic scene understanding through scene flow estimation and object segmentation","abstract":"Understanding dynamic 3D scenes—critical for applications like autonomous navigation and mixed reality—requires pars- ing both motion (scene flow) and object interactions (segmen- tation). Scene flow captures 3D motion fields, while segmen- tation isolates objects, enabling systems to interpret evolving environments. Integrating these tasks offers a holistic view but faces computational challenges due to scene flow’s high dimensionality. This work proposes a lightweight deep learning architecture combining an enhanced Point Transformer for efficient fea- ture extraction and a point-voxel correlation module for sta- ble motion estimation. To bypass labor-intensive object annotations, scene flow is leveraged as auxiliary supervision. Instead of predicting masks for all points, this thesis focuses on key points, reducing com- plexity while maintaining accuracy. The proposed clustering- free approach achieves state-of-the-art results on indoor datasets. For temporal consistency, an unsupervised method integrates continuous point cloud sequences (encoding spatial embed- dings) with time-independent queries (encoding object se- mantics). This enables gradual mask prediction across frames without direct labels, accommodating dynamic inputs. This framework advances dynamic scene understanding by harmo- nizing motion and segmentation, validated through competi- tive benchmarks and flexible input handling.","abstract_html":"Understanding dynamic 3D scenes—critical for applications like autonomous navigation and mixed reality—requires pars- ing both motion (scene flow) and object interactions (segmen- tation). Scene flow captures 3D motion fields, while segmen- tation isolates objects, enabling systems to interpret evolving environments. Integrating these tasks offers a holistic view but faces computational challenges due to scene flow’s high dimensionality. This work proposes a lightweight deep learning architecture combining an enhanced Point Transformer for efficient fea- ture extraction and a point-voxel correlation module for sta- ble motion estimation. To bypass labor-intensive object annotations, scene flow is leveraged as auxiliary supervision. Instead of predicting masks for all points, this thesis focuses on key points, reducing com- plexity while maintaining accuracy. The proposed clustering- free approach achieves state-of-the-art results on indoor datasets. For temporal consistency, an unsupervised method integrates continuous point cloud sequences (encoding spatial embed- dings) with time-independent queries (encoding object se- mantics). This enables gradual mask prediction across frames without direct labels, accommodating dynamic inputs. This framework advances dynamic scene understanding by harmo- nizing motion and segmentation, validated through competi- tive benchmarks and flexible input handling.","abstract_has_math":false,"creators":["Li, Zhiqi"],"institution":"Bournemouth University","degree_name":null,"degree_level":"doctoral","degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2025,"date_issued":"2025-04","date_published":"2025-04","updated_at":"2026-07-24T01:12:54Z","subjects":[],"languages":["en"],"rights":[],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":null,"outbound_label":null,"outbound_source":null},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:creator","label":"Author","values":["Li, Zhiqi"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2025-04-29"]},{"key":"dc:date.issued","label":"Date","values":["2025-04"]},{"key":"dc:publisher.department","label":"Dc Publisher Department","values":["Faculty of Media and Communication"]},{"key":"dc:publisher.institution","label":"Dc Publisher Institution","values":["Bournemouth University"]},{"key":"dc:relation.isreferencedby","label":"Dc Relation Isreferencedby","values":["https://eprints.bournemouth.ac.uk/41016/"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]},{"key":"dc:type.qualificationlevel","label":"Dc Type Qualificationlevel","values":["doctoral"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["en"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://eprints.bournemouth.ac.uk/41016/1/LI%2C%20Zhiqi_Ph.D._2025.pdf"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Understanding dynamic 3D scenes—critical for applications like autonomous navigation and mixed reality—requires pars- ing both motion (scene flow) and object interactions (segmen- tation). Scene flow captures 3D motion fields, while segmen- tation isolates objects, enabling systems to interpret evolving environments. Integrating these tasks offers a holistic view but faces computational challenges due to scene flow’s high dimensionality. This work proposes a lightweight deep learning architecture combining an enhanced Point Transformer for efficient fea- ture extraction and a point-voxel correlation module for sta- ble motion estimation. To bypass labor-intensive object annotations, scene flow is leveraged as auxiliary supervision. Instead of predicting masks for all points, this thesis focuses on key points, reducing com- plexity while maintaining accuracy. The proposed clustering- free approach achieves state-of-the-art results on indoor datasets. For temporal consistency, an unsupervised method integrates continuous point cloud sequences (encoding spatial embed- dings) with time-independent queries (encoding object se- mantics). This enables gradual mask prediction across frames without direct labels, accommodating dynamic inputs. This framework advances dynamic scene understanding by harmo- nizing motion and segmentation, validated through competi- tive benchmarks and flexible input handling."]},{"key":"dc:format","label":"Dc Format","values":["application/pdf"]},{"key":"dc:title","label":"Title","values":["Empower dynamic scene understanding through scene flow estimation and object segmentation"]}]}],"canonical_facts":{"dc:creator":["Li, Zhiqi"],"dc:date":["2025-04-29"],"dc:date.issued":["2025-04"],"dc:description.abstract":["Understanding dynamic 3D scenes—critical for applications like autonomous navigation and mixed reality—requires pars- ing both motion (scene flow) and object interactions (segmen- tation). Scene flow captures 3D motion fields, while segmen- tation isolates objects, enabling systems to interpret evolving environments. Integrating these tasks offers a holistic view but faces computational challenges due to scene flow’s high dimensionality. This work proposes a lightweight deep learning architecture combining an enhanced Point Transformer for efficient fea- ture extraction and a point-voxel correlation module for sta- ble motion estimation. To bypass labor-intensive object annotations, scene flow is leveraged as auxiliary supervision. Instead of predicting masks for all points, this thesis focuses on key points, reducing com- plexity while maintaining accuracy. The proposed clustering- free approach achieves state-of-the-art results on indoor datasets. For temporal consistency, an unsupervised method integrates continuous point cloud sequences (encoding spatial embed- dings) with time-independent queries (encoding object se- mantics). This enables gradual mask prediction across frames without direct labels, accommodating dynamic inputs. This framework advances dynamic scene understanding by harmo- nizing motion and segmentation, validated through competi- tive benchmarks and flexible input handling."],"dc:format":["application/pdf"],"dc:identifier.uri":["https://eprints.bournemouth.ac.uk/41016/1/LI%2C%20Zhiqi_Ph.D._2025.pdf"],"dc:language":["en"],"dc:publisher.department":["Faculty of Media and Communication"],"dc:publisher.institution":["Bournemouth University"],"dc:relation.isreferencedby":["https://eprints.bournemouth.ac.uk/41016/"],"dc:title":["Empower dynamic scene understanding through scene flow estimation and object segmentation"],"dc:type":["Thesis"],"dc:type.qualificationlevel":["doctoral"]},"updated_at":"2026-07-24T01:12:54Z"}