Visual-Inertial Odometry (VIO) is a widely used state estimation technique for Uncrewed Aerial Vehicle (UAV) navigation in environments where Global Navigation Satellite System (GNSS) signals are unavailable. VIO systems that rely on visual feature tracking are susceptible to performance degradation when operating over surfaces containing repetitive visual textures, where visually similar features can produce ambiguous correspondences that introduce errors into the trajectory estimate. Despite the prevalence of repetitive textures in indoor UAV operating environments such as warehouses, manufacturing facilities, and infrastructure corridors, the specific impact of different repetitive pattern geometries on per-surface VIO accuracy has received limited systematic study, and to the best of the author's knowledge, the relationship between feature-level metrics extracted from the camera stream and per-surface VIO trajectory error has not been previously characterized on the UAV platform class evaluated here with motion capture ground truth.
This thesis presents a systematic comparison of feature-level metrics and VIO trajectory accuracy across three repetitive floor patterns (horizontal stripes, diamond lattice, and houndstooth) and a non-repetitive bare-floor control surface, characterizing both pattern-induced trajectory error and the relationship between feature-level metric values and per-surface trajectory accuracy. Experiments were conducted using a ModalAI Sentinel UAV equipped with the VOXL2 flight computer running the onboard Qualcomm Visual-Inertial Odometry (qVIO) system, with Vicon motion capture providing ground truth reference data across twelve trials. Fourteen feature-level metrics were computed using descriptor-based Oriented FAST and Rotated BRIEF (ORB) and motion-based Shi-Tomasi with Kanade-Lucas-Tomasi (KLT) tracking. The analysis was applied externally as offline post-processing on the recorded camera frames, without modification of the proprietary qVIO pipeline. The metrics included feature count, spatial distribution, descriptor uniqueness through Local Uniqueness Score (LUS), temporal consistency through Temporal Consistency Score (TCS), and tracking persistence. The bare-floor control produced a pooled Root Mean Square Error (RMSE) of 3.26 cm with consistent per-trial accuracy. The three repetitive pattern conditions produced pooled RMSE values of 6.05 to 6.74 cm with substantially greater per-trial variability, corresponding to pattern-to-baseline ratios of approximately 1.9 to 2.1. Cross-detector validation found that all twelve rank assignments across surfaces were identical between ORB and Shi-Tomasi supporting the interpretation that the per-surface property rankings reflect surface properties rather than detector-configuration artifacts. None of the fourteen feature-level metrics reproduced the per-surface trajectory error ordering observed for the tested VIO implementation. Spearman rank correlation analysis across the twelve trials produced no correlation reaching statistical significance at the conventional threshold. The repetitive-pattern conditions produced higher trajectory RMSE than the bare-floor control, while the RMSE differences among the three repetitive patterns were comparatively small. This decoupling between feature-level metric values and integrated trajectory accuracy may inform the design of texture-aware VIO systems, and the results provide a quantitative reference for indoor UAV operations on repetitive surfaces.
","abstract_html":"<p>Visual-Inertial Odometry (VIO) is a widely used state estimation technique for Uncrewed Aerial Vehicle (UAV) navigation in environments where Global Navigation Satellite System (GNSS) signals are unavailable. VIO systems that rely on visual feature tracking are susceptible to performance degradation when operating over surfaces containing repetitive visual textures, where visually similar features can produce ambiguous correspondences that introduce errors into the trajectory estimate. Despite the prevalence of repetitive textures in indoor UAV operating environments such as warehouses, manufacturing facilities, and infrastructure corridors, the specific impact of different repetitive pattern geometries on per-surface VIO accuracy has received limited systematic study, and to the best of the author's knowledge, the relationship between feature-level metrics extracted from the camera stream and per-surface VIO trajectory error has not been previously characterized on the UAV platform class evaluated here with motion capture ground truth.</p> <p>This thesis presents a systematic comparison of feature-level metrics and VIO trajectory accuracy across three repetitive floor patterns (horizontal stripes, diamond lattice, and houndstooth) and a non-repetitive bare-floor control surface, characterizing both pattern-induced trajectory error and the relationship between feature-level metric values and per-surface trajectory accuracy. Experiments were conducted using a ModalAI Sentinel UAV equipped with the VOXL2 flight computer running the onboard Qualcomm Visual-Inertial Odometry (qVIO) system, with Vicon motion capture providing ground truth reference data across twelve trials. Fourteen feature-level metrics were computed using descriptor-based Oriented FAST and Rotated BRIEF (ORB) and motion-based Shi-Tomasi with Kanade-Lucas-Tomasi (KLT) tracking. The analysis was applied externally as offline post-processing on the recorded camera frames, without modification of the proprietary qVIO pipeline. The metrics included feature count, spatial distribution, descriptor uniqueness through Local Uniqueness Score (LUS), temporal consistency through Temporal Consistency Score (TCS), and tracking persistence. The bare-floor control produced a pooled Root Mean Square Error (RMSE) of 3.26 cm with consistent per-trial accuracy. The three repetitive pattern conditions produced pooled RMSE values of 6.05 to 6.74 cm with substantially greater per-trial variability, corresponding to pattern-to-baseline ratios of approximately 1.9 to 2.1. Cross-detector validation found that all twelve rank assignments across surfaces were identical between ORB and Shi-Tomasi supporting the interpretation that the per-surface property rankings reflect surface properties rather than detector-configuration artifacts. None of the fourteen feature-level metrics reproduced the per-surface trajectory error ordering observed for the tested VIO implementation. Spearman rank correlation analysis across the twelve trials produced no correlation reaching statistical significance at the conventional threshold. The repetitive-pattern conditions produced higher trajectory RMSE than the bare-floor control, while the RMSE differences among the three repetitive patterns were comparatively small. This decoupling between feature-level metric values and integrated trajectory accuracy may inform the design of texture-aware VIO systems, and the results provide a quantitative reference for indoor UAV operations on repetitive surfaces.<strong></strong></p>","abstract_has_math":false,"creators":["El Mekkoussi, Anass"],"institution":null,"degree_name":"Master of Science in Unmanned and Autonomous Systems Engineering","degree_level":"Thesis - Open Access","degree_discipline":"Electrical Engineering and Computer Science","degree_department":null,"school":null,"contributors":[],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2026,"date_issued":"2026-04-01T07:00:00Z","date_published":"2026-04-01T07:00:00Z","updated_at":"2026-07-27T19:26:22Z","subjects":["visual-inertial odometry","qVIO","repetitive patterns","ORB","Shi-Tomasi","feature detection","UAV","GNSS-denied navigation","trajectory error","Vicon ground truth","Aerospace Engineering","Aviation","Electrical and Computer Engineering","Mechanical Engineering","Operations Research, Systems Engineering and Industrial Engineering"],"languages":[],"rights":[],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://commons.erau.edu/edt/998","outbound_label":"Repository record","outbound_source":"dc:identifier"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:creator","label":"Author","values":["El Mekkoussi, Anass"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"thesis:degree_discipline","label":"Discipline","values":["Electrical Engineering and Computer Science"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Thesis - Open Access"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Master of Science in Unmanned and Autonomous Systems Engineering"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["visual-inertial odometry","qVIO","repetitive patterns","ORB","Shi-Tomasi","feature detection","UAV","GNSS-denied navigation","trajectory error","Vicon ground truth","Aerospace Engineering","Aviation","Electrical and Computer Engineering","Mechanical Engineering","Operations Research, Systems Engineering and Industrial Engineering"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["https://commons.erau.edu/edt/998"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Visual-Inertial Odometry (VIO) is a widely used state estimation technique for Uncrewed Aerial Vehicle (UAV) navigation in environments where Global Navigation Satellite System (GNSS) signals are unavailable. VIO systems that rely on visual feature tracking are susceptible to performance degradation when operating over surfaces containing repetitive visual textures, where visually similar features can produce ambiguous correspondences that introduce errors into the trajectory estimate. Despite the prevalence of repetitive textures in indoor UAV operating environments such as warehouses, manufacturing facilities, and infrastructure corridors, the specific impact of different repetitive pattern geometries on per-surface VIO accuracy has received limited systematic study, and to the best of the author's knowledge, the relationship between feature-level metrics extracted from the camera stream and per-surface VIO trajectory error has not been previously characterized on the UAV platform class evaluated here with motion capture ground truth.
This thesis presents a systematic comparison of feature-level metrics and VIO trajectory accuracy across three repetitive floor patterns (horizontal stripes, diamond lattice, and houndstooth) and a non-repetitive bare-floor control surface, characterizing both pattern-induced trajectory error and the relationship between feature-level metric values and per-surface trajectory accuracy. Experiments were conducted using a ModalAI Sentinel UAV equipped with the VOXL2 flight computer running the onboard Qualcomm Visual-Inertial Odometry (qVIO) system, with Vicon motion capture providing ground truth reference data across twelve trials. Fourteen feature-level metrics were computed using descriptor-based Oriented FAST and Rotated BRIEF (ORB) and motion-based Shi-Tomasi with Kanade-Lucas-Tomasi (KLT) tracking. The analysis was applied externally as offline post-processing on the recorded camera frames, without modification of the proprietary qVIO pipeline. The metrics included feature count, spatial distribution, descriptor uniqueness through Local Uniqueness Score (LUS), temporal consistency through Temporal Consistency Score (TCS), and tracking persistence. The bare-floor control produced a pooled Root Mean Square Error (RMSE) of 3.26 cm with consistent per-trial accuracy. The three repetitive pattern conditions produced pooled RMSE values of 6.05 to 6.74 cm with substantially greater per-trial variability, corresponding to pattern-to-baseline ratios of approximately 1.9 to 2.1. Cross-detector validation found that all twelve rank assignments across surfaces were identical between ORB and Shi-Tomasi supporting the interpretation that the per-surface property rankings reflect surface properties rather than detector-configuration artifacts. None of the fourteen feature-level metrics reproduced the per-surface trajectory error ordering observed for the tested VIO implementation. Spearman rank correlation analysis across the twelve trials produced no correlation reaching statistical significance at the conventional threshold. The repetitive-pattern conditions produced higher trajectory RMSE than the bare-floor control, while the RMSE differences among the three repetitive patterns were comparatively small. This decoupling between feature-level metric values and integrated trajectory accuracy may inform the design of texture-aware VIO systems, and the results provide a quantitative reference for indoor UAV operations on repetitive surfaces.
"]},{"key":"dc:title","label":"Title","values":["Evaluating UAV Visual-Inertial Odometry Trajectory Error and Feature-Level Metrics over Repetitive Floor Patterns"]}]}],"canonical_facts":{"dc:creator":["El Mekkoussi, Anass"],"dc:description.abstract":["Visual-Inertial Odometry (VIO) is a widely used state estimation technique for Uncrewed Aerial Vehicle (UAV) navigation in environments where Global Navigation Satellite System (GNSS) signals are unavailable. VIO systems that rely on visual feature tracking are susceptible to performance degradation when operating over surfaces containing repetitive visual textures, where visually similar features can produce ambiguous correspondences that introduce errors into the trajectory estimate. Despite the prevalence of repetitive textures in indoor UAV operating environments such as warehouses, manufacturing facilities, and infrastructure corridors, the specific impact of different repetitive pattern geometries on per-surface VIO accuracy has received limited systematic study, and to the best of the author's knowledge, the relationship between feature-level metrics extracted from the camera stream and per-surface VIO trajectory error has not been previously characterized on the UAV platform class evaluated here with motion capture ground truth.
This thesis presents a systematic comparison of feature-level metrics and VIO trajectory accuracy across three repetitive floor patterns (horizontal stripes, diamond lattice, and houndstooth) and a non-repetitive bare-floor control surface, characterizing both pattern-induced trajectory error and the relationship between feature-level metric values and per-surface trajectory accuracy. Experiments were conducted using a ModalAI Sentinel UAV equipped with the VOXL2 flight computer running the onboard Qualcomm Visual-Inertial Odometry (qVIO) system, with Vicon motion capture providing ground truth reference data across twelve trials. Fourteen feature-level metrics were computed using descriptor-based Oriented FAST and Rotated BRIEF (ORB) and motion-based Shi-Tomasi with Kanade-Lucas-Tomasi (KLT) tracking. The analysis was applied externally as offline post-processing on the recorded camera frames, without modification of the proprietary qVIO pipeline. The metrics included feature count, spatial distribution, descriptor uniqueness through Local Uniqueness Score (LUS), temporal consistency through Temporal Consistency Score (TCS), and tracking persistence. The bare-floor control produced a pooled Root Mean Square Error (RMSE) of 3.26 cm with consistent per-trial accuracy. The three repetitive pattern conditions produced pooled RMSE values of 6.05 to 6.74 cm with substantially greater per-trial variability, corresponding to pattern-to-baseline ratios of approximately 1.9 to 2.1. Cross-detector validation found that all twelve rank assignments across surfaces were identical between ORB and Shi-Tomasi supporting the interpretation that the per-surface property rankings reflect surface properties rather than detector-configuration artifacts. None of the fourteen feature-level metrics reproduced the per-surface trajectory error ordering observed for the tested VIO implementation. Spearman rank correlation analysis across the twelve trials produced no correlation reaching statistical significance at the conventional threshold. The repetitive-pattern conditions produced higher trajectory RMSE than the bare-floor control, while the RMSE differences among the three repetitive patterns were comparatively small. This decoupling between feature-level metric values and integrated trajectory accuracy may inform the design of texture-aware VIO systems, and the results provide a quantitative reference for indoor UAV operations on repetitive surfaces.
"],"dc:identifier":["https://commons.erau.edu/edt/998"],"dc:subject":["visual-inertial odometry","qVIO","repetitive patterns","ORB","Shi-Tomasi","feature detection","UAV","GNSS-denied navigation","trajectory error","Vicon ground truth","Aerospace Engineering","Aviation","Electrical and Computer Engineering","Mechanical Engineering","Operations Research, Systems Engineering and Industrial Engineering"],"dc:title":["Evaluating UAV Visual-Inertial Odometry Trajectory Error and Feature-Level Metrics over Repetitive Floor Patterns"],"thesis:degree_discipline":["Electrical Engineering and Computer Science"],"thesis:degree_level":["Thesis - Open Access"],"thesis:degree_name":["Master of Science in Unmanned and Autonomous Systems Engineering"]},"updated_at":"2026-07-27T19:26:22Z"}