Ronan, William, Wang, Zhaokai et Wang, Xian.
2025.
« Quantifying aleatoric uncertainty for mobile robot object detection ».
In Proceedings of the CSME-CFDSC-CSR 2025 International Congress (Montreal, QC, Canada, May 25-28, 2025)
Coll. « Progress in Canadian Mechanical Engineering », vol. 8.
Prévisualisation |
PDF
185 - Quantifying aleatoric uncertainty .pdf - Version publiée Licence d'utilisation : Tous les droits réservés aux détenteurs du droit d'auteur. Télécharger (2MB) | Prévisualisation |
Résumé
Object detection in computer vision is inherently associated with uncertainty, which can be categorized into two primary types: aleatoric uncertainty and epistemic uncertainty. Aleatoric uncertainty represents the inherent randomness of real-world data and usually cannot be reduced. In the context of mobile robot perception, aleatoric uncertainty is prevalent in images with poor quality, compromised visibility, and increased visual complexity. These factors can significantly degrade object detection performance, thereby impairing robotic decision-making and functionality. Further, the mobile nature of the robots introduces increased aleatoric uncertainty due to variation in image backgrounds, motion blur, and other environmental influences. This study proposes a methodology to predict the effect of aleatoric uncertainty on object detection performance in real-time. Using a Quanser QCar experimental setup, we derived aleatoric metrics from image data - contrast, edge, brightness, velocity - to quantify uncertainty factors latent within the image. Prediction of the object detection algorithm’s performance is conducted using supervised learning to train models on the relationship between our aleatoric metrics and object detection ability. The results demonstrate the ability to preemptively assess the limitations of object detection under varying environmental conditions, providing a predictive framework for improving the reliability and safety of mobile robotic systems.
| Type de document: | Compte rendu de conférence |
|---|---|
| Éditeurs: | Éditeurs ORCID Hof, Lucas A. NON SPÉCIFIÉ Di Labbio, Giuseppe NON SPÉCIFIÉ Tahan, Antoine NON SPÉCIFIÉ Sanjosé, Marlène NON SPÉCIFIÉ Lalonde, Sébastien NON SPÉCIFIÉ Demarquette, Nicole R. NON SPÉCIFIÉ |
| Date de dépôt: | 18 déc. 2025 15:32 |
| Dernière modification: | 18 déc. 2025 16:47 |
| URI: | https://espace2.etsmtl.ca/id/eprint/32491 |
Actions (Authentification requise)
 |
Dernière vérification avant le dépôt |