3D Pot-Shaped Object Localization Based on RGB-Image and Point Clouds

Abstract

Existing network-based 3D object detection and localization methods require a processing platform with high computing power. When applied to industrial scenarios with low computing power and low power consumption, they cannot meet the real-time requirements. In this paper, we propose a novel 3D pot-shaped object localization method based on the RGB-image and point clouds. To meet the requirement of speed and accuracy, we adopt the idea of extracting features in images and locating the object in point clouds. Firstly, we extract object region and edge information from high-resolution images, and the edge point clouds are generated based on edge information. Then we design a pot-shaped object localization method based on space plane constraint, which converts complex 3D circle fitting to 2D plane circle and reduces the time-consuming greatly. Experiments demonstrate that our method achieves a better trade-off between accuracy and speed. And compared with others, our method also has superiority.