PointNet Based Point Cloud Segmentation of Steel Plate Blank Stack
The estimation of the layer thickness of the steel plate blank during the process of disassembling and stacking is crucial for the accurate and safe execution of the pushing action by the steel pushing machine.At present,many enterprises still rely on manual observation by operators to estimate the thickness of steel plates and the relative height with the conveyor roller in this production process.Inaccurate estimation can easily lead to collisions,resulting in damage to the conveyor equipment and production interruption.An intelligent layering method for steel plate blank stacks is proposed.The method combines the on-site working environment with laser radar to perform three-dimensional point cloud imaging of the steel plate blank stack.Then,the collected point cloud data is subjected to feature recognition,layered segmentation,and extraction using the PointNet neural network framework.Finally,different segmented layers are converted into real thickness based on calibration values.According to the on-site experimental results,PointNet achieves a recognition rate of 87.4%for steel plate blank stack segmentation,with a thickness estimation error of less than 1.2 cm.Based on the steel plate blank specification table(three specifications of 150,160,and 180 cm),the thickness specification of the steel plate blank can be accurately estimated,with a recognition rate is 15f/s,meeting the requirements of on-site working conditions.
万方数据
维普
