Monocular 3D Pedestrian Detection and Tracking With Neural Network with Uncertainty Estimate for Safe Drone Operation. Flegel, T., Praveen Jawaharlal Ayyanathan, Ehsan Taheri, & Bevly, D. In January, 2025. ![Monocular 3D Pedestrian Detection and Tracking With Neural Network with Uncertainty Estimate for Safe Drone Operation [link]](https://bibbase.org/img/filetypes/link.svg "link")
Paper abstract bibtex In order for drone use to become more integrated in daily life, drones need to be able to accurately detect and localize humans in their surroundings. We propose a solution to this problem using a monocular camera. A notable feature of the proposed algorithm is its independence from joint detection or depth estimation algorithms, which are more computationally complex. These features make it particularly suited for embedded systems and robotics applications on resource-constrained platforms, including drones. In particular, a neural network based approach is utilized to accurately detect and localize humans by estimating their relative position, and velocity, as well as an associated uncertainty of the estimates. Our results indicate that the position estimate is comparable to other methods in the literature, despite being a lighter weight algorithm, with the velocity estimates showing the overall velocity trend, but requiring further improvements with respect to accuracy.
@inproceedings{flegel_monocular_2025,
title = {Monocular {3D} {Pedestrian} {Detection} and {Tracking} {With} {Neural} {Network} with {Uncertainty} {Estimate} for {Safe} {Drone} {Operation}},
url = {https://doi.org/10.2514/6.2025-2117},
abstract = {In order for drone use to become more integrated in daily life, drones need to be able to accurately detect and localize humans in their surroundings. We propose a solution to this problem using a monocular camera. A notable feature of the proposed algorithm is its independence from joint detection or depth estimation algorithms, which are more computationally complex. These features make it particularly suited for embedded systems and robotics applications on resource-constrained platforms, including drones. In particular, a neural network based approach is utilized to accurately detect and localize humans by estimating their relative position, and velocity, as well as an associated uncertainty of the estimates. Our results indicate that the position estimate is comparable to other methods in the literature, despite being a lighter weight algorithm, with the velocity estimates showing the overall velocity trend, but requiring further improvements with respect to accuracy.},
author = {Flegel, Tyler and {Praveen Jawaharlal Ayyanathan} and {Ehsan Taheri} and Bevly, David},
month = jan,
year = {2025},
}
Downloads: 0
{"_id":"ab8ZFLyeamo3NSBgC","bibbaseid":"flegel-praveenjawaharlalayyanathan-ehsantaheri-bevly-monocular3dpedestriandetectionandtrackingwithneuralnetworkwithuncertaintyestimateforsafedroneoperation-2025","author_short":["Flegel, T.","Praveen Jawaharlal Ayyanathan","Ehsan Taheri","Bevly, D."],"bibdata":{"bibtype":"inproceedings","type":"inproceedings","title":"Monocular 3D Pedestrian Detection and Tracking With Neural Network with Uncertainty Estimate for Safe Drone Operation","url":"https://doi.org/10.2514/6.2025-2117","abstract":"In order for drone use to become more integrated in daily life, drones need to be able to accurately detect and localize humans in their surroundings. We propose a solution to this problem using a monocular camera. A notable feature of the proposed algorithm is its independence from joint detection or depth estimation algorithms, which are more computationally complex. These features make it particularly suited for embedded systems and robotics applications on resource-constrained platforms, including drones. In particular, a neural network based approach is utilized to accurately detect and localize humans by estimating their relative position, and velocity, as well as an associated uncertainty of the estimates. Our results indicate that the position estimate is comparable to other methods in the literature, despite being a lighter weight algorithm, with the velocity estimates showing the overall velocity trend, but requiring further improvements with respect to accuracy.","author":[{"propositions":[],"lastnames":["Flegel"],"firstnames":["Tyler"],"suffixes":[]},{"firstnames":[],"propositions":[],"lastnames":["Praveen Jawaharlal Ayyanathan"],"suffixes":[]},{"firstnames":[],"propositions":[],"lastnames":["Ehsan Taheri"],"suffixes":[]},{"propositions":[],"lastnames":["Bevly"],"firstnames":["David"],"suffixes":[]}],"month":"January","year":"2025","bibtex":"@inproceedings{flegel_monocular_2025,\n\ttitle = {Monocular {3D} {Pedestrian} {Detection} and {Tracking} {With} {Neural} {Network} with {Uncertainty} {Estimate} for {Safe} {Drone} {Operation}},\n\turl = {https://doi.org/10.2514/6.2025-2117},\n\tabstract = {In order for drone use to become more integrated in daily life, drones need to be able to accurately detect and localize humans in their surroundings. We propose a solution to this problem using a monocular camera. A notable feature of the proposed algorithm is its independence from joint detection or depth estimation algorithms, which are more computationally complex. These features make it particularly suited for embedded systems and robotics applications on resource-constrained platforms, including drones. In particular, a neural network based approach is utilized to accurately detect and localize humans by estimating their relative position, and velocity, as well as an associated uncertainty of the estimates. Our results indicate that the position estimate is comparable to other methods in the literature, despite being a lighter weight algorithm, with the velocity estimates showing the overall velocity trend, but requiring further improvements with respect to accuracy.},\n\tauthor = {Flegel, Tyler and {Praveen Jawaharlal Ayyanathan} and {Ehsan Taheri} and Bevly, David},\n\tmonth = jan,\n\tyear = {2025},\n}\n\n\n\n","author_short":["Flegel, T.","Praveen Jawaharlal Ayyanathan","Ehsan Taheri","Bevly, D."],"key":"flegel_monocular_2025","id":"flegel_monocular_2025","bibbaseid":"flegel-praveenjawaharlalayyanathan-ehsantaheri-bevly-monocular3dpedestriandetectionandtrackingwithneuralnetworkwithuncertaintyestimateforsafedroneoperation-2025","role":"author","urls":{"Paper":"https://doi.org/10.2514/6.2025-2117"},"metadata":{"authorlinks":{}}},"bibtype":"inproceedings","biburl":"https://bibbase.org/zotero-group/keb0115/5574615","dataSources":["kDK6fZ4EDThxNKDCP"],"keywords":[],"search_terms":["monocular","pedestrian","detection","tracking","neural","network","uncertainty","estimate","safe","drone","operation","flegel","praveen jawaharlal ayyanathan","ehsan taheri","bevly"],"title":"Monocular 3D Pedestrian Detection and Tracking With Neural Network with Uncertainty Estimate for Safe Drone Operation","year":2025}