Accuracy Study of Close Range 3D Object Reconstruction Based on Point Clouds

The paper presents the results of multi-view imagery 3D reconstruction of the close range building using the geodetic and photogrammetric networks and image based point clouds. DSLR camera was used to acquire the block of 28 terrestrial multiple convergent photos. To create the benchmark data set, the bundle block adjustment with camera self calibration in the Pictran software package was performed. For the 3D reconstruction the RealityCapture, Photoscan and Pix4D software were used. Software comparison was performed using mesh models. The analysis was concerned the comparison of processing parameters, checking the visual quality of the outputs and the evaluation of the 3D models accuracy. The study proved the highest usefulness of the RealityCapture application. The tested software allowed to the point matching on the control points with mean standard deviation lower than 0.5 pixel, mean adjustment RMS deviation lower than 3 mm and average points position error lower than 4 mm.