Deep learning-based object detection with point cloud data

Abstract

Deep convolutional neural networks (CNNs) are used in various tasks, especially in classification and object detection in two-dimensional images. In this work, two deep convolutional neural networks were experimented for detecting objects from three-dimensional point cloud data. Neural network models can utilize the additional depth information of point clouds to learn spatial features based on the locations of the data points. Using deep convolutional neural networks for object detection has promising results but creating a point cloud dataset from scratch requires time. The aim of the experi-ments was to create an own dataset that fits the pre-defined models. The dataset had only several examples for experimenting the models, but good datasets should be as large as possible. The point clouds need complex processing to ensure effectiveness and precise accuracy for the model.