{"id":{"repo_id":"usf","oai_identifier":"oai:digitalcommons.usf.edu:etd-10995"},"canonical_url":"https://search.ndltd.org/etd/usf/oai:digitalcommons.usf.edu:etd-10995","repository":{"repo_id":"usf","name":"University of South Florida","base_url":"https://digitalcommons.usf.edu/do/oai/"},"display":{"title":"A Study of Deep Learning Silhouette Extractors for Gait Recognition","abstract":"Gait recognition has gained importance in biometrics for identifying a person. It is performed by analyzing the gait features extracted from a video of a walking person. Gait features can be represented using silhouettes, pose elements, 3d human body features, or the combination of any of the above features. Most gait recognition algorithms use silhouette data for modeling. Therefore the quality of the silhouettes is crucial for achieving better recognition results. Many previous works have used the silhouettes given with the dataset for developing a gait model. In this work, we will test the performance of different background subtraction algorithms on the gait recognition problem. We compared background subtraction methods: Background Matting, Hybrid Task Cascade, Mixture of Gaussians, Fully Convolutional Networks, and DeeplabV3. We performed the experiments on CASIA-B and USF HumanID datasets, which were collected indoors and outdoors, respectively, using static cameras. The casiaB dataset considers changes in two covariates: view angle and walking surface, and USF HumanID dataset considers changes in five covariates: change in view angle, change in the walking surface, change in shoe type, with or without the briefcase, and change in time. This variation in the datasets would give us an idea of how background subtraction methods perform under various conditions. The efficacy of these silhouette extractors is evaluated on five state-of-the-art gait recognition algorithms: GaitSet, GaitPart, GLN, OpenGait Baseline, and GEI-Net by comparing Rank-1 identification accuracies, and their statistical significance is shown by two-way ANOVA analysis. We observed that the silhouettes generated by matting had achieved a minimum of 1.34% increase in all the experiments.","abstract_html":"Gait recognition has gained importance in biometrics for identifying a person. It is performed by analyzing the gait features extracted from a video of a walking person. Gait features can be represented using silhouettes, pose elements, 3d human body features, or the combination of any of the above features. Most gait recognition algorithms use silhouette data for modeling. Therefore the quality of the silhouettes is crucial for achieving better recognition results. Many previous works have used the silhouettes given with the dataset for developing a gait model. In this work, we will test the performance of different background subtraction algorithms on the gait recognition problem. We compared background subtraction methods: Background Matting, Hybrid Task Cascade, Mixture of Gaussians, Fully Convolutional Networks, and DeeplabV3. We performed the experiments on CASIA-B and USF HumanID datasets, which were collected indoors and outdoors, respectively, using static cameras. The casiaB dataset considers changes in two covariates: view angle and walking surface, and USF HumanID dataset considers changes in five covariates: change in view angle, change in the walking surface, change in shoe type, with or without the briefcase, and change in time. This variation in the datasets would give us an idea of how background subtraction methods perform under various conditions. The efficacy of these silhouette extractors is evaluated on five state-of-the-art gait recognition algorithms: GaitSet, GaitPart, GLN, OpenGait Baseline, and GEI-Net by comparing Rank-1 identification accuracies, and their statistical significance is shown by two-way ANOVA analysis. We observed that the silhouettes generated by matting had achieved a minimum of 1.34% increase in all the experiments.","abstract_has_math":false,"creators":["Oladhri, Sneha"],"institution":"University of South Florida","degree_name":"MS in Computer Science (M.S.C.S.)","degree_level":"Thesis","degree_discipline":"Computer Science and Engineering","degree_department":null,"school":null,"contributors":["Sudeep Sarkar, Ph.D.","Mauricio Pamplona Segundo, Ph.D.","Shaun Canavan, Ph.D."],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2022,"date_issued":"2022-10-31T07:00:00Z","date_published":"2022-10-31T07:00:00Z","updated_at":"2026-11-19T08:48:36Z","subjects":["Background Subtraction","Background Matting","Segmentation","Statistical Analysis","ANOVA","Engineering"],"languages":[],"rights":[],"rights_urls":[],"identifier_entries":[],"person_identifier_entries":[]},"links":{"outbound_url":"https://digitalcommons.usf.edu/etd/9798","outbound_label":"Repository record","outbound_source":"dc:identifier","record_url":"https://digitalcommons.usf.edu/etd/9798","record_label":"Repository record","record_kind":"record","record_source":"dc:identifier","download_url":null,"download_label":null,"download_source":null,"access_status":null,"embargo_until":null,"access_note":null},"source_record":{"url":"https://digitalcommons.usf.edu/do/oai/?verb=GetRecord&metadataPrefix=oai_etdms&identifier=oai%3Adigitalcommons.usf.edu%3Aetd-10995","prefix":"oai_etdms"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Sudeep Sarkar, Ph.D.","Mauricio Pamplona Segundo, Ph.D.","Shaun Canavan, Ph.D."]},{"key":"dc:creator","label":"Author","values":["Oladhri, Sneha"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2022-10-31T07:00:00Z"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Computer Science and Engineering"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Thesis","M.S.C.S."]},{"key":"thesis:degree_name","label":"Degree Name","values":["MS in Computer Science (M.S.C.S.)"]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["University of South Florida"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Background Subtraction","Background Matting","Segmentation","Statistical Analysis","ANOVA","Engineering"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["https://digitalcommons.usf.edu/etd/9798"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Gait recognition has gained importance in biometrics for identifying a person. It is performed by analyzing the gait features extracted from a video of a walking person. Gait features can be represented using silhouettes, pose elements, 3d human body features, or the combination of any of the above features. Most gait recognition algorithms use silhouette data for modeling. Therefore the quality of the silhouettes is crucial for achieving better recognition results. Many previous works have used the silhouettes given with the dataset for developing a gait model. In this work, we will test the performance of different background subtraction algorithms on the gait recognition problem. We compared background subtraction methods: Background Matting, Hybrid Task Cascade, Mixture of Gaussians, Fully Convolutional Networks, and DeeplabV3. We performed the experiments on CASIA-B and USF HumanID datasets, which were collected indoors and outdoors, respectively, using static cameras. The casiaB dataset considers changes in two covariates: view angle and walking surface, and USF HumanID dataset considers changes in five covariates: change in view angle, change in the walking surface, change in shoe type, with or without the briefcase, and change in time. This variation in the datasets would give us an idea of how background subtraction methods perform under various conditions. The efficacy of these silhouette extractors is evaluated on five state-of-the-art gait recognition algorithms: GaitSet, GaitPart, GLN, OpenGait Baseline, and GEI-Net by comparing Rank-1 identification accuracies, and their statistical significance is shown by two-way ANOVA analysis. We observed that the silhouettes generated by matting had achieved a minimum of 1.34% increase in all the experiments."]},{"key":"dc:title","label":"Title","values":["A Study of Deep Learning Silhouette Extractors for Gait Recognition"]}]}],"canonical_facts":{"dc:contributor":["Sudeep Sarkar, Ph.D.","Mauricio Pamplona Segundo, Ph.D.","Shaun Canavan, Ph.D."],"dc:creator":["Oladhri, Sneha"],"dc:date":["2022-10-31T07:00:00Z"],"dc:description":["Gait recognition has gained importance in biometrics for identifying a person. It is performed by analyzing the gait features extracted from a video of a walking person. Gait features can be represented using silhouettes, pose elements, 3d human body features, or the combination of any of the above features. Most gait recognition algorithms use silhouette data for modeling. Therefore the quality of the silhouettes is crucial for achieving better recognition results. Many previous works have used the silhouettes given with the dataset for developing a gait model. In this work, we will test the performance of different background subtraction algorithms on the gait recognition problem. We compared background subtraction methods: Background Matting, Hybrid Task Cascade, Mixture of Gaussians, Fully Convolutional Networks, and DeeplabV3. We performed the experiments on CASIA-B and USF HumanID datasets, which were collected indoors and outdoors, respectively, using static cameras. The casiaB dataset considers changes in two covariates: view angle and walking surface, and USF HumanID dataset considers changes in five covariates: change in view angle, change in the walking surface, change in shoe type, with or without the briefcase, and change in time. This variation in the datasets would give us an idea of how background subtraction methods perform under various conditions. The efficacy of these silhouette extractors is evaluated on five state-of-the-art gait recognition algorithms: GaitSet, GaitPart, GLN, OpenGait Baseline, and GEI-Net by comparing Rank-1 identification accuracies, and their statistical significance is shown by two-way ANOVA analysis. We observed that the silhouettes generated by matting had achieved a minimum of 1.34% increase in all the experiments."],"dc:identifier":["https://digitalcommons.usf.edu/etd/9798"],"dc:subject":["Background Subtraction","Background Matting","Segmentation","Statistical Analysis","ANOVA","Engineering"],"dc:title":["A Study of Deep Learning Silhouette Extractors for Gait Recognition"],"thesis:degree_discipline":["Computer Science and Engineering"],"thesis:degree_level":["Thesis","M.S.C.S."],"thesis:degree_name":["MS in Computer Science (M.S.C.S.)"],"thesis:institution_name":["University of South Florida"]},"updated_at":"2026-11-19T08:48:36Z"}