LiDAR-Based Detection of Urban Trees Using a Backpack System

Abstract. Mapping individual trees accurately in densely populated urban environments is challenging due to occlusion effects, overlapping canopies, and irregular tree morphology. This paper presents and evaluates an automated tree detection technique based on the vertical continuity principle to decrease the reliance on preprocessing steps, such as terrain filtering and point cloud normalization. A heuristic filter effectively distinguishes trees from pole-like structures that demonstrate vertical continuity, which helps to reduce false positives. Data were collected using a backpack LiDAR (Light Detection and Ranging) system with an accuracy of 5 cm. The sensor’s effective range is up to 50 m (at 80% reflectivity), enabling the acquisition of high-density point clouds at close-range distances while maintaining efficiency and accessibility in complex urban environments. The method was tested across three diverse urban sites, with 156 trees, and achieved an F_Score of 94.1%, with a 26 cm horizontal RMSEXY in trunk positioning.