Research On Point Cloud Data Processing Technology Of Special Shaped Workpiece Based On Depth Image

With the advent of the intelligent era,all industries are trying to develop intelligently.Automatic spraying occupies an important field in the industrial field,which is widely used in aviation,military affairs,ships and other fields,and improves the spraying efficiency.Although the intelligent level of spraying robot has been improved,which can replace most manual spraying,at present,intelligent spraying still uses manual teaching method to plan the spraying path,which not only affects the safety of workers,but also affects the spraying quality.Therefore,it is urgent to study the automatic spraying technology,that is,to plan the path by offline programming and then let the robot plan its own path for automatic spraying.In this paper,aiming at the point cloud data obtained from industrial field,combined with deep learning method,the point cloud data collected from industrial field are processed,and the automatic spraying of spraying robot is studied and analyzed.Firstly,the theoretical basis and network structure of deep learning network models Pointnet and Pointnet++are introduced.Training deep learning network models Pointnet and Pointnet++by processing their own point cloud data,and starting the path planning of spraying process according to the classification results.This paper introduces some operations of point cloud data processing,including removing outliers of point cloud by using direct filtering and statistical filtering,downsampling point cloud data by using voxel grid method,simplifying point cloud data,calculating normal vector of point cloud and reconstructing surface.Then,according to the main influencing factors of spraying engineering,the coating thickness deposition model of spraying process is established.The objective optimization function of coating thickness distribution uniformity is established,the constraint conditions and single objective are established,and the optimization parameters are solved to prepare for the subsequent spraying path planning.Finally,according to the previously processed point cloud data and the obtained optimal solution parameters,the spraying path points are obtained by using the point cloud slicing method,and the final spraying path is obtained by sorting the path points.At the end of the experiment,the algorithm is realized according to the algorithm flow,and the experimental results are analyzed.At first,the deep learning framework PyTorch 1.1.0 is built under Ubuntu 16.04 operating system to complete the classification experiment of point cloud data.Follow-up experiments were carried out according to the accuracy.The classification result is over 90%,so the point cloud data is processed and the rendering is displayed.The final spraying path is obtained according to the processed point cloud data and the optimized parameters.The rendering shows that the obtained path of the abnormal point cloud data is relatively complete.