AUTOMATIC POINT CLOUD NOISE MASKING IN CLOSE RANGE PHOTOGRAMMETRY FOR BUILDINGS USING AI-BASED SEMANTIC LABELLING

Abstract. The use of AI in semantic segmentation has grown significantly recent years, aided by developments computing power and the availability annotated images for training data. However, context close-range photogrammetry, although working with 2D images, is still used mostly 3D point cloud purposes. In this paper, we propose a simple method to apply such methods close range photogrammetry benefitting from deep learning-based segmentation. Specifically, was detect unwanted objects scene involving reconstruction historical building façade. For these purposes, classes e.g., sky, trees, electricity poles were considered as noise. Masks then created results which would constraint dense image matching process only wanted classes. regard, resulting essentially projected labels into space, thus excluding noise object scene. Our compared manual masking managed achieve comparable while requiring fraction processing time when using pre-trained DL network do task.