ABILITY ANALYSIS OF THE POINT-TO-PLANE METHOD ON ITERATIVE CLOSEST POINT (ICP) ALGORITHM IN INCREASING ACCURACY RESULTS OF THE EARTH^S SURFACE POINT CLOUD STRIP ADJUSTMENT
Monica Maharani(a*), Riyas Syamsul Arif,(a) Harintaka(b)

a) Universitas Pembangunan Nasional Veteran Yogyakarta
b) Universitas Gadjah Mada, Yogyakarta

Abstract

Point-to-plane in the Iterative Closest Point (ICP) algorithm is a registration
method in 3D Machine Learning that is reliable in increasing the accuracy of
registration of non-topographic objects. Research is necessary when this method is applied to topographic objects which have different characteristics from nontopographic objects, especially in the variety of objects. This study was conducted to examine the reliability of the point-to-plane method for registering point cloud LiDAR on real-world topographic objects as the implementation of strip adjustment. This research begins with building a program to perform strip adjustment. The program uses Iterative Closest Point (ICP) algorithm with a point-to-plane registration method. The program is developed on Phycharm as an Integrated Development Environment (IDE) using Python Language. Then, the program is used to perform a strip adjustment process on 2 scene point cloud data from the acquisition of an Unmanned Aerial Vehicle with a Light Detection and Ranging (LiDAR) sensor. The result of the strip adjustment on the program is compared with the strip adjustment result using the point-to-point method on CloudCompare software. The comparison is to determine the reliability and accuracy of the registration result from the point-to plane method through the value of the Root Mean Square Error (RMSE), transformation matrices, fitness, correspondence, and visual appearance. In addition, the comparison is also done on each type of land cover on the earth^s surface, like roof, vegetation, roads, and ground. This step is done to test the performance of the point to-plane method on each type of land cover on topographic data. In this study, it is proven that the point-to-plane registration method has better accuracy than the point-to-point in registering point cloud data on the earth^s surface. The RMSE value using the point-to-point method is 8.71 centimeters, while the point to-plane method is 1.53 centimeters. The RMSE value and transformation matrix on
the point-top-plane are more minor than point-to-point. The point-to-plane method produces a fitness value of 0.001 meters with a total correspondence of 1054 points. For each type of land cover on the earth^s surface, the point-to-plane method has succeeded in making fitness on the combination of data.

Keywords: point-to-plane, strip adjustment, ICP, machine learning

Topic: Solid Earth Sciences