Performance Analysis of Iterative Closest Point (ICP) Algorithm using Modified Hausdorff Distance

1School of Electronics Engineering, VIT University, Vellore, India 2Assistant Professor (Selection Grade), VIT University, Vellore, India ---------------------------------------------------------------------***--------------------------------------------------------------------Abstract Registration, tracking and reconstruction of 3-D models in real time has many applications in the fields of pose estimation, alignment, motion estimation, object recognition, handheld scanning and augmented reality. Many varied techniques and approaches have been used to tackle the problem of registration and reconstruction of 3-D images. Our work aims to design a method for accurate and computationally efficient registration of 3-D shapes, through comparison and application of various alignment techniques. The expected outcome is a system able to align and reconstruct geometrically precise complex 3-D models in a fast and efficient manner. We mainly focus on the Iterative Closest Point Algorithm, which is the most popular method used in this field through many different forms and variations. An effort has been made to analyze these various forms which are used for object registration through the use of Modified Hausdorff Distance (MHD). Usage of outlier removal methods such as Random Sampling Consensus (RANSAC) and accelerating techniques such as kd-Tree search are also studied. This system could act as a foundation for implementing the ICP algorithm along with depth images or Simultaneous Localization and Mapping (SLAM).