The objective of this coursework is to generate a 3D model of the inside of an office room using a set of 3D point clouds generated by scanning the room using an Intel RealSense depth sensor. The original data provided consists of 40 frames taken from different viewpoints as the sensor is moved across the office. One clear pattern that can be noticed by examining the frames is that frames include similar scenes to one another (especially consecutive frames) so this pattern will be crucial in the reconstruction process. The work can be divided into the following tasks:
- Extract the relevant data from each point cloud
- Estimate the reference transformation linking each consecutive pair of frames
- Fuse all of the points into a single 3D coordinate system.
- Evaluate the quality of the final model
A set of relatively robust algorithms for performing several cleaning actions on both pointcloud and RGB data associated with depth range data recorded from the inside of an office was generated. Moreover, an algorithm for merging consecutive frames based on matching SIFT points was proposed.