Statistical assessment for point cloud dataset

Abstract

Geospatial data is used in geomatics or land surveying. It provides geographical information and spatial primitives in the measured objects. Geospatial data can be developed into semantic values with digitalisation, including software and automation. For example, point clouds generated by laser scanners and photogrammetry can be rendered into a 3D model. Rendering is effective for the high richness of details of the object (e.g. buildings) compared to the point clouds, which contain noise and inconsistency of details. However, how significantly the 3D model can improve the disadvantage of the point clouds shall be determined. This paper presents a statistical analysis using hypothesis testing to identify either point clouds or 3D models are significantly different in geometry. Two (2) hypotheses were outlined; null hypothesis and alternate hypothesis between both samples data where significant difference and no significant difference were determined respectively. The statistic calculation was done using twenty (20) sample data extracted from each dataset. Root mean square error (RMSE) shows that point clouds and the 3D model are 0.132 and 0.455, respectively, evaluating that the point cloud is more accurate in geometry than the 3D model. At-test was performed to calculate the probability value. The result shows the probability value of 0.97 is greater than the alpha value of 0.05. Hence, the null hypothesis is failed to reject. This statistic test shows that the point clouds and 3D models are not significantly different in geometry.