3D Point Cloud Classification Algorithm based on Improved PointNet++

3D point cloud classification is one of the important tasks before autonomous driving technology. However, the traditional PointNet++ algorithm has some limitations, such as sensitivity to noise and outliers, and high computational complexity. To solve these problems, this paper proposes a 3D point cloud classification algorithm based on improved PointNet++. This algorithm adds a layer of statistical filtering method to preprocess the input point cloud data before inputting the point cloud into the PointNet++ network, which effectively improves the quality of the point cloud. At the same time, the CONV1D convolution is changed to a depth-separable convolution, which reduces the computational complexity. Experimental results show that compared with the traditional PointNet++ algorithm, our improved model can achieve better training accuracy and training efficiency under limited training resources.