Electronic Thesis and Dissertation Repository


Master of Science




Wang, Jinfei


Previous work has identified a positive relationship between the density of aerial LiDAR input for building reconstruction and the accuracy of the resulting reconstructed models. We hypothesize a point of diminished returns at which higher data density no longer contributes meaningfully to higher accuracy in the end product. We investigate this relationship by subsampling a high-density dataset from the City of Surrey, BC to different densities and inputting each subsampled dataset to reconstruction using two different reconstruction methods. We then determine the accuracy of reconstruction based on manually created reference data, in terms of both 2D footprint accuracy and 3D model accuracy. We find that there is no quantitative evidence for meaningfully improved output accuracy from densities higher than 4 p/m2 for either method, although aesthetic improvements at higher point cloud densities are noted for one method.