A Coal Flow Point Cloud Sampling and Segmentation Algorithm for Belt Conveyors Based on Improved Graph Laplacian Operator

Binocular vision-based point cloud reconstruction is pivotal for real-time detection of coal flow on belt conveyors. It plays a critical role in ensuring the safe and energy-efficient operation of such conveyors. However, during binocular vision detection, the coal flow point cloud exhibits highly complex morphology, and coal lump edges remain notoriously challenging to segment. These issues significantly degrade the accuracy and real-time performance of point cloud reconstruction. To address this issue, a coal flow point cloud sampling and segmentation method for belt conveyors based on an improved graph Laplacian operator is proposed. Specifically, the captured coal flow point cloud is processed using a locally adaptive sampling method integrated with an information entropy optimization function. Meanwhile, a Laplace-PointNN model is constructed to extract features from the coal flow point cloud and establish a feature library. These steps enable precise segmentation of point clouds at the edges of coal lumps, thereby facilitating the completion of coal flow point cloud reconstruction. Experimental results show that the proposed method has an average relative error of 3.48%, and an 84.5% increase in detection speed with an average running time of 0.58 seconds. The results demonstrate that the algorithm can realize real-time detection of coal flow on belt conveyors, ensuring their safe and energy-efficient operation.