Comparison of Semantic Segmentation of Point Clouds Obtained from Different Sensors Using Deep Learning

Deep learning methods have been successfully used in image processing and computer vision. Point cloud semantic segmentation is also a current study subject where deep learning is widely used. In this study, Semantic3D, a terrestrial laser scanning data, and Dublin City, an airborne laser scanning data, were used. Random sampling and an effective local feature aggregator (RANDLA-Net) were used as the segmentation algorithm. Precision, recall, F1 score, and overall accuracy were used as evaluation metrics. The overall accuracy is obtained as 0.882 in the Semantic3D dataset and 0.896 in the Dublin City dataset.