Disturbance Analysis in the Classification of Objects Obtained from Urban LiDAR Point Clouds with Convolutional Neural Networks

Mobile Laser Scanning (MLS) systems have proven their usefulness in the rapid and accurate acquisition of urban environment. From generated point clouds, street furniture can be extracted classified without manual intervention. However, this process classification is not error-free, caused mainly by disturbances. This paper analyses effect three disturbances (point density variation, ambient noise, occlusions) on objects clouds. clouds acquired real case studies, synthetic are added. The reduction downsampling a voxel-wise distribution. noise as random points within bounding box object, occlusion eliminating contained sphere. Samples with pre-trained Convolutional Neural Network (CNN). results showed different behaviours for each disturbance: affected depending object shape dimensions, volume while occlusions depended size location. Finally, CNN was re-trained percentage samples An improvement performance 10–40% reported except radius larger than 1 m.