Monocular Vision-Based Stockpile Localization in Waste Bunkers

Résumé fourni par la source

Monocular vision localization is a preferred noncontact solution for specific scenarios due to its low cost and flexible deployment. This paper proposes a monocular recognition and localization method for waste bunkers stockpiles. The optimal HSV range is determined through sample calibration; dual denoising combining Gaussian blur and morphological erosion as well as threshold segmentation with contour filtering ensure accurate recognition. A pinhole imaging-based model is built, with Kalman filter mitigating motion blur, and camera calibration establishing pixel-physical size mapping. Large models including Doubao, Grok and Qwen tested for target size estimation, providing insights for complex tasks such as monocular depth estimation. This method, with over$\mathbf{8 0 \%}$hardware cost reduction compared with lidar and industrial-standard performance, offers a cost-effective solution for automated waste bunkers handling.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.