Research Problem
Existing point cloud generation tools in Blender work by creating mesh models of virtual scenes and extract random points from surfaces, edges or mesh nodes. This does not reflect well how points clouds are created from really existing physical LIDAR scanners such as the rotating scanner VLP-16. They may look sensible ton the human eye, but to train machine learning algorithms to work correctly for real-world data, it is important that we can generate ground truth data that is entirely realistic but based on well-know virtual scenes.
The Thesis
The literature review of this thesis explores relevant studies and existing techniques for point cloud generation from virtual scenes. It focuses on modern methods and algorithms used in computer graphics and computer vision to generate realistic point clouds. Additionally, it examines previous attempts to integrate LIDAR scanner data with Blender, highlighting the challenges encountered and the gaps that this research aims to address.
The research objectives are as follows:
- Develop a method to simulate the VLP-16 rotating LIDAR scanner within Blender, accurately capturing the geometry and reflectance properties of virtual scenes.
- Implement algorithms for transforming the captured data into point cloud representations, considering the characteristics and specifications of the VLP-16 rotating LIDAR scanner.
- Evaluate the quality and accuracy of the generated point clouds by comparing them with ground truth data and established benchmark datasets.
- Explore methods for optimizing the point cloud generation process in terms of computational efficiency and scalability.
The thesis will describe the steps involved in achieving the research objectives. It outlines the process of simulating the VLP-16 rotating LIDAR scanner within Blender, including capturing the geometry, reflectance, and positioning information of virtual scenes. Algorithms for transforming the captured data into point clouds will be implemented, considering the scanner's specifications and characteristics. The evaluation phase will involve quantitative analysis by comparing the generated point clouds with ground truth data and established benchmarks. Additionally, optimization techniques will be explored to improve the efficiency and scalability of the point cloud generation process.
Potential Contributions
This research aims to contribute to the field of point cloud generation by providing a modern and functional solution for generating accurate point clouds from virtual scenes in Blender using the VLP-16 rotating LIDAR scanner. The proposed solution will enable researchers and practitioners to generate realistic point clouds for various applications, including robotics, simulation, and virtual reality. Furthermore, this study will contribute to the advancement of point cloud generation algorithms and serve as a foundation for future research on integrating other LIDAR scanner data with Blender or similar software.
The software should be released on Github.
Learning outcome
Experience in
- in formulating, investigating and answering research questions
- understanding LiDARs and their properties and error models
- knowing about 3D modeling in general and in particular in Blender
Conditions
We expect that you:
- have been admitted to a master's program in MatNat@UiO - primarily PROSA
- take this as a long thesis
- will participate actively in the weekly SINLab meetings
- are present in the lab and collaborate with other students and staff
- are interested in and have some knowledge of Python and C/C++ programming
- are willing to share your results on Github
- include the course IN5060 in the study plan, unless you have already completed a course on classical (non-ML) data analysis