Road-Side Individual Tree Segmentation from Urban MLS Point Clouds Using Metric Learning

As one of the most important components urban space, an outdated inventory road-side trees may misguide managers in assessment and upgrade environments, potentially affecting road quality. Therefore, automatic accurate instance segmentation from point clouds is task ecology research. However, previous works show under- or over-segmentation effects for due to overlapping, irregular shapes incompleteness. In this paper, a deep learning framework that combines semantic proposed extract single vehicle-mounted mobile laser scanning (MLS) clouds. stage, ground points are filtered reduce processing time. Subsequently, graph-based network developed segment tree raw MLS For individual novel joint adopted detect instance-level roadside trees. Two complex Chinese cloud scenes used evaluate performance method. The method accurately approximately 90% achieve better results than existing published methods both two Living Vegetation Volume (LVV) calculation can benefit segmentation. provides promising solution ecological construction based on LVV roads.