{"id":{"repo_id":"texas-state","oai_identifier":"oai:digital.library.txst.edu:10877/24790"},"canonical_url":"https://search.dev.ndltd.org/etd/texas-state/oai:digital.library.txst.edu:10877/24790","repository":{"repo_id":"texas-state","name":"Texas State University","base_url":"https://digital.library.txst.edu/server/oai/request"},"display":{"title":"Neural Region-of-Interest-Aware Video Compression for Wildlife Monitoring Under Edge Computing Constraints","abstract":"Wildlife camera traps can produce far more video than is practical to store locally or transmit from remote sites. In many clips, the animal occupies only a small part of the frame, while most of the image is background. Standard whole-frame compression does not account for this imbalance and may allocate bitrate to regions that matter less for later review. This thesis presents a region-of-interest (ROI)-aware video compression methodology for wildlife footage that achieves substantially higher compression ratios than conventional methods. It first identifies the animal region on selected frames and tracks it between detections. It then compresses the animal region and the background in separate sparse streams with different compression settings. The compressed video binaries, along with the metadata required for video reconstruction, are then archived and transmitted as a ZIP file to the cloud server. The evaluation used region-specific and whole-frame quality measures, transmitted-archive size, runtime analysis, and an edge device deployment study on a Jetson Nano Orin. In the main controlled study, the released DCVC-RT neural codec configuration reduced the transmitted archives from 399.16 MB of source video to 10.36 MB across the 20 held-out clips, corresponding to a 97.40% reduction in transmitted archive size, while preserving the animal region more faithfully than the frame as a whole. The deployment study further showed that the proposed split-stream, ROI-aware video compression methodology is codec-agnostic. The same design can be implemented using traditional codecs, such as HEVC and AV1, depending on hardware acceleration capabilities.","abstract_html":"Wildlife camera traps can produce far more video than is practical to store locally or transmit from remote sites. In many clips, the animal occupies only a small part of the frame, while most of the image is background. Standard whole-frame compression does not account for this imbalance and may allocate bitrate to regions that matter less for later review. This thesis presents a region-of-interest (ROI)-aware video compression methodology for wildlife footage that achieves substantially higher compression ratios than conventional methods. It first identifies the animal region on selected frames and tracks it between detections. It then compresses the animal region and the background in separate sparse streams with different compression settings. The compressed video binaries, along with the metadata required for video reconstruction, are then archived and transmitted as a ZIP file to the cloud server. The evaluation used region-specific and whole-frame quality measures, transmitted-archive size, runtime analysis, and an edge device deployment study on a Jetson Nano Orin. In the main controlled study, the released DCVC-RT neural codec configuration reduced the transmitted archives from 399.16 MB of source video to 10.36 MB across the 20 held-out clips, corresponding to a 97.40% reduction in transmitted archive size, while preserving the animal region more faithfully than the frame as a whole. The deployment study further showed that the proposed split-stream, ROI-aware video compression methodology is codec-agnostic. The same design can be implemented using traditional codecs, such as HEVC and AV1, depending on hardware acceleration capabilities.","abstract_has_math":false,"creators":["Mathew, Felix"],"institution":"Texas State University","degree_name":"Master of Science","degree_level":"Masters","degree_discipline":"Computer Science","degree_department":null,"school":null,"contributors":[],"advisors":["Metsis, Vangelis"],"committee_chairs":[],"committee_members":["Ngu, Anne H. H.","Farias, Mylene Queiroz de"],"year":2026,"date_issued":"2026-05","date_published":"2026-05","updated_at":"2026-07-27T21:22:34Z","subjects":["wildlife monitoring","camera traps","region-of-interest (ROI)","neural video compression","edge computing"],"languages":["en"],"rights":[],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://hdl.handle.net/10877/24790","outbound_label":"Handle","outbound_source":"dc:identifier.uri"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Metsis, Vangelis"]},{"key":"dc:contributor.committeemember","label":"Committee Member","values":["Ngu, Anne H. H.","Farias, Mylene Queiroz de"]},{"key":"dc:creator","label":"Author","values":["Mathew, Felix"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2026-05-11T20:44:49Z"]},{"key":"dc:date.issued","label":"Date","values":["2026-05"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Computer Science"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Masters"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Master of Science"]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["Texas State University"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["wildlife monitoring","camera traps","region-of-interest (ROI)","neural video compression","edge computing"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["en"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://hdl.handle.net/10877/24790"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Wildlife camera traps can produce far more video than is practical to store locally or transmit from remote sites. In many clips, the animal occupies only a small part of the frame, while most of the image is background. Standard whole-frame compression does not account for this imbalance and may allocate bitrate to regions that matter less for later review. This thesis presents a region-of-interest (ROI)-aware video compression methodology for wildlife footage that achieves substantially higher compression ratios than conventional methods. It first identifies the animal region on selected frames and tracks it between detections. It then compresses the animal region and the background in separate sparse streams with different compression settings. The compressed video binaries, along with the metadata required for video reconstruction, are then archived and transmitted as a ZIP file to the cloud server. The evaluation used region-specific and whole-frame quality measures, transmitted-archive size, runtime analysis, and an edge device deployment study on a Jetson Nano Orin. In the main controlled study, the released DCVC-RT neural codec configuration reduced the transmitted archives from 399.16 MB of source video to 10.36 MB across the 20 held-out clips, corresponding to a 97.40% reduction in transmitted archive size, while preserving the animal region more faithfully than the frame as a whole. The deployment study further showed that the proposed split-stream, ROI-aware video compression methodology is codec-agnostic. The same design can be implemented using traditional codecs, such as HEVC and AV1, depending on hardware acceleration capabilities."]},{"key":"dc:format","label":"Dc Format","values":["Text"]},{"key":"dc:format.medium","label":"Dc Format Medium","values":["1 file (.pdf)"]},{"key":"dc:title","label":"Title","values":["Neural Region-of-Interest-Aware Video Compression for Wildlife Monitoring Under Edge Computing Constraints"]}]}],"canonical_facts":{"dc:contributor.advisor":["Metsis, Vangelis"],"dc:contributor.committeemember":["Ngu, Anne H. H.","Farias, Mylene Queiroz de"],"dc:creator":["Mathew, Felix"],"dc:date.accessioned":["2026-05-11T20:44:49Z"],"dc:date.issued":["2026-05"],"dc:description.abstract":["Wildlife camera traps can produce far more video than is practical to store locally or transmit from remote sites. In many clips, the animal occupies only a small part of the frame, while most of the image is background. Standard whole-frame compression does not account for this imbalance and may allocate bitrate to regions that matter less for later review. This thesis presents a region-of-interest (ROI)-aware video compression methodology for wildlife footage that achieves substantially higher compression ratios than conventional methods. It first identifies the animal region on selected frames and tracks it between detections. It then compresses the animal region and the background in separate sparse streams with different compression settings. The compressed video binaries, along with the metadata required for video reconstruction, are then archived and transmitted as a ZIP file to the cloud server. The evaluation used region-specific and whole-frame quality measures, transmitted-archive size, runtime analysis, and an edge device deployment study on a Jetson Nano Orin. In the main controlled study, the released DCVC-RT neural codec configuration reduced the transmitted archives from 399.16 MB of source video to 10.36 MB across the 20 held-out clips, corresponding to a 97.40% reduction in transmitted archive size, while preserving the animal region more faithfully than the frame as a whole. The deployment study further showed that the proposed split-stream, ROI-aware video compression methodology is codec-agnostic. The same design can be implemented using traditional codecs, such as HEVC and AV1, depending on hardware acceleration capabilities."],"dc:format":["Text"],"dc:format.medium":["1 file (.pdf)"],"dc:identifier.uri":["https://hdl.handle.net/10877/24790"],"dc:language.iso":["en"],"dc:subject":["wildlife monitoring","camera traps","region-of-interest (ROI)","neural video compression","edge computing"],"dc:title":["Neural Region-of-Interest-Aware Video Compression for Wildlife Monitoring Under Edge Computing Constraints"],"dc:type":["Thesis"],"thesis:degree_discipline":["Computer Science"],"thesis:degree_level":["Masters"],"thesis:degree_name":["Master of Science"],"thesis:institution_name":["Texas State University"]},"updated_at":"2026-07-27T21:22:34Z"}