3D reconstruction and visualization

Abstract

The project report is about obtaining 3D reconstruction based on 2D images. In the context of 3D models and visualization of environments, reconstruction systems from unordered image collection face the issue of robustness, accuracy, completeness, and scalability during the process of generating fully satisfactory results. However, tremendous advancements in many aspects experienced over the years, the increasing importance for a number of current and future application areas are undeniable advantages of 3D representations. So the goal is to illustrate the first steps, which are essential to construct a 3D environment, to visualize the results in order to identify the most appropriate algorithm for a 3D model of three objects having diverse textures. The taken approach is derived from state-of-the-art literature.

As final experimental outcomes, the used method suggests that it is feasible to recalculate correction in the sparse point of cloud by using bundle adjustment.

Introduction

Existing reconstruction structures for three-dimensional physical objects are of great importance in numerous current and future application fields, such as content creation for computer games and geospatial applications, 3D walkthroughs of real estates, digital preservation of cultural heritage and crime site, and so on. Due to tremendous advancements in recent years, a wide variety of approaches for digitization and reconstruction on the shapes of 3D physical objects.

Problem statement

Physical objects are continually created from digital models day-to-day in many industries,

including engineering, building construction, manufacturing . Although, the reverse problem,

an inference of a digital representation from an existing physical object or environment hasn`t gained much more attention. The problem is regarded as a reverse-engineering or 3D reconstruction. There are several features of 3D environment that appear to be interested in

recovery processes, such as its shape, color, material, and even its functional properties.

A large quantity of range samples and pictures are required to achieve an environment model. As a consequence, mass of data demand algorithms that can enable to reproduce computer models from those great samples efficiently and reliably. As a final step, in order to complete reconstruction, the physical object needs to be scanned from different viewpoints which turns out to be challenging. The most appropriate views and the required quantity of them need to be identified for covering an entire scene.

Some problems, such as how to arrange the geometry of environment surface with the color cameras, how to transform all the data into a one coordinate system reliably, how to convert detached scans into a single surface description, how to merge the color representations related to the diverse scenes to achieve an efficient texture reconstruction should be addressed in order to reach reconstruction process of textured 3D models.

The overall procedure to carry out the project can be divided into five major phases:

1. Exploring SFM , SIFT algorithms, models and approaches.
2. Researching related work and implementations.
3. Developing the approach for image processing relied on the findings.
4. Implementing the approach and drafting the concept.
5. Evaluating and discussing the outcomes.