This paper presents a comprehensive survey of recent progress in deep learning methods for 3D point clouds, covering three major tasks: 3D shape classification, 3D object detection and tracking, and 3D point cloud segmentation. It also provides comparative results on several publicly available datasets, along with insightful observations and future research directions. The paper discusses various methods for each task, including multi-view based, volumetric-based, and point-based approaches. For 3D shape classification, methods typically learn point embeddings and extract global shape embeddings. For 3D object detection and tracking, methods include convolution-based, graph-based, and hierarchical data structure-based approaches. For 3D point cloud segmentation, methods include semantic segmentation, instance segmentation, and part segmentation. The paper also discusses the challenges of deep learning on 3D point clouds, such as small dataset scale, high dimensionality, and unstructured nature. It highlights the importance of permutation invariance and the use of various architectures like PointNet, PointNet++, and graph convolution networks. The paper also discusses the use of different evaluation metrics for each task, including overall accuracy, mean class accuracy, average precision, and mean intersection over union. The paper concludes that deep learning on 3D point clouds is a rapidly evolving field with many promising research directions.This paper presents a comprehensive survey of recent progress in deep learning methods for 3D point clouds, covering three major tasks: 3D shape classification, 3D object detection and tracking, and 3D point cloud segmentation. It also provides comparative results on several publicly available datasets, along with insightful observations and future research directions. The paper discusses various methods for each task, including multi-view based, volumetric-based, and point-based approaches. For 3D shape classification, methods typically learn point embeddings and extract global shape embeddings. For 3D object detection and tracking, methods include convolution-based, graph-based, and hierarchical data structure-based approaches. For 3D point cloud segmentation, methods include semantic segmentation, instance segmentation, and part segmentation. The paper also discusses the challenges of deep learning on 3D point clouds, such as small dataset scale, high dimensionality, and unstructured nature. It highlights the importance of permutation invariance and the use of various architectures like PointNet, PointNet++, and graph convolution networks. The paper also discusses the use of different evaluation metrics for each task, including overall accuracy, mean class accuracy, average precision, and mean intersection over union. The paper concludes that deep learning on 3D point clouds is a rapidly evolving field with many promising research directions.