indoor-reconstruction

This repository uses Git Large File Storage because of the data set that is supplied. This extension can be found at: https://git-lfs.github.com. To pull the data set, run git lfs pull.

This reconstruction system is developed by us, Fredrik Kristoffer Johanssen and Vetle Smedbakken Sillerud, as part of our master thesis in Cybernetics; "Reconstruction of Indoor Environments Using LiDAR and IMU." Below is a reconstruction of both authors; Fredrik on the left, and Vetle on the right.

Summary

The goal of this thesis is to develop a method for reconstructing accurate point cloud representations of indoor environments using the Velodyne LiDAR Puck 16 (VLP-16) and the Tinkerforge IMU 2.0. The following points can summarize the general goals of this thesis:

Develop a data acquisition system for collecting LiDAR and IMU data. This system must handle synchronization of the data acquisition in order to prepare the scans for alignment.
Generate high-density point clouds by combining LiDAR and IMU data.

We conclude that at subset level, our reconstruction system can reconstruct high-density point clouds of indoor environments with a precision that is mostly limited to the inherent uncertainties of the VLP- 16. We also conclude that the registration of several subsets obtained from different positions is able to preserve both visual appearance and reflective intensity of objects in the scene. Our reconstruction system can thus be utilized to generate real data sets of high-density point clouds.

System setup

The system setup consists of; Tinkerforge IMU 2.0, 3D printed IMU mounting cap, VLP-16 LiDAR, Hama Star 61 153-3D tripod, VLP-16 Interface Box, Computer.

Developed method

Our developed method consists of two parts: data acquisition and data processing. The data acquisition is written in C++ and can be found in the C++ folder. The data processing is written in python and can be found in the python folder.

Results

Result from one stationary position:

Results from several positions with odometry estimate between each:

The coordinate frames, in red, green and blue, indicates the position where a subset is generated. The circles with arrows indicate the position where a snapshot of the reconstructed environment is taken, along with the an image of the actual scene from this position. The arrow indicate in which direction the snapshot is captured. The orange line indicate the translation trajectory travelled during this acquisition.

Datasets

We supply one data set in this repository due to space limitations. Other data sets can be provided upon request. We divide one LiDAR scan, i.e., acquisition of points in one instance of the LiDAR acquisition scope, into data packets to handle movement over time better. The data sets are structured in the following manner:

All data is contained in the main directory. The main directory is named with the name of the data set, e.g. "poster", as supplied in this repository.
Subdirectory "fragments" contains all the data of the subsets, i.e. fragments, of the environment. These are named by order they are acquired, i.e., "fragment0" are acquired before "fragment1".
Subdirectory "odometry" contains all the data of needed to estimate the translation between the fragments. These are also named by the order of which they are acquired.
The data contained in the subdirectories of these subdirectories are the acquired orientation measurements and point clouds, as well as the generated pose graphs and combined clouds.
"quaternions_datapacket.csv" contains the quaternions associated with the individual data packets.
"quaternions_interpolated.csv" contains the quaternions that are generated from "quaternions_datapacket.csv" after linear interpolation.
"quaternions_fragment.csv" contains the quaternions associated with the full scans, i.e., the data packets that are combined to LiDAR scan.
"imu_data.csv" contains the angular velocity and the linear accelerations in x, y and z directions, associated with each data packet.
E.g., "scan_2_5.pcd" is the point cloud of the 5th data packet in 2nd instance of a LiDAR scan.
E.g., "scan2.pcd" is the point cloud that is the combination of all data packets in the 2nd instance of a LiDAR scan.
"combined_cloud.pcd" is the point cloud that is yielded by the combination of all instance of LiDAR scans. These scans are only transformed using the IMU orientation measurements.
"multiway_registration.pcd" is the point cloud that is yielded by the registration of all fragments, i.e., the registration of all "combined_clouds.pcd" contained in subdirectories of the "fragments" directory.
"pose_graph.json" contains the pose graph that is generated in the registration process.

vetlesillerud / indoor-reconstruction Goto Github PK