Point cloud recognition of street tree canopies in urban Internet of Things based on laser reflection intensity

Abstract

Light Detection and Ranging (LiDAR) technology, as a core component of the IoT perception layer, has become a research focus for street tree canopy target recognition. However, traditional methods relying on point cloud geometric features often struggle to achieve accurate identification in complex scenarios where tree canopies intertwine with adjacent objects. To address this issue, this study proposes a novel point cloud recognition method based on laser reflection intensity. First, a 2D LiDAR combined with Mobile Laser Scanning (MLS) technology was employed to collect training datasets (distance-intensity and incidence angle-intensity) for constructing an intensity correction model. Subsequently, urban street point cloud intensity data were acquired using a 2D LiDAR-based MLS system, followed by distance and incidence angle correction. Finally, the intensity threshold for canopy recognition was determined based on the probability density distribution of the corrected intensity data. To validate the method’s effectiveness, the intensity threshold calibrated from a 40-meter road segment was applied to another 80-meter segment within the same street scene. The performance of the original and corrected intensity thresholds was then compared. Experimental results demonstrated that the corrected intensity threshold achieved an F1-score of 0.84 for canopy point cloud recognition, representing a 31 % improvement over the original threshold (F1-score: 0.64). This confirms that the proposed method significantly enhances recognition accuracy in complex urban environments.