wangpengcv / kinectfusion Goto Github PK

The goal of this project is to obtain a 3D reconstruction of a static scene which was recorded with a Kinect camera. The Kinect camera provides RGB as well as depth information as can be seen in the picture below.

Below you can see a result of a raycast which was performed after ~120 frames.

First clone the repo with git clone https://github.com/VladimirYugay/KinectFusion.git.

• Install cmake if it's not installed yet
  sudo apt-get install cmake

• Run sudo apt-get install libfreeimage3 libfreeimage-dev in order to be able to process images from the dataset by simulating the virtal sensor

• Download header only Eigen library and put it into libs folder by runnning:
  cd KinectFusion/libs
git clone https://gitlab.com/libeigen/eigen.git
  Use any stable release (e.g. 3.4.0) as the upstream currently requires specific compiler versions (newer than GCC 9.3.0 and Intel 2021.4.0.20210910)
  cd eigen
git checkout tags/3.4.0

• Follow the same file structure as in Linux Setup
• Add pre-compiled libraries of FreeImage in KinectFusion/libs

• brew install eigen
• brew install freeimage

You can either use your own image sequence which was recorded with a Kinect Camera v2 or you can download the same dataset we used here. Extract the folder and move it to KinectFusion/data/rgbd_dataset_freiburg1_xyz.

In order to build and run the project follow these steps:

cd ..
mkdir build
cd build
cmake ..
make
./kinect_fusion

If you want to contribute, please use cpplint. You can install it via pip install cpplint and run with: cpplint [OPTIONS] files.

You can find GPU implementation on vy/unstable/gpu_integration and vy/icp/gpu_icp_pose_estimation branches

Things implemented on CUDA

1. Finding correspondences
2. ICP linear system construction vy/icp/gpu_icp_pose_estimation
3. Volume integration