Skip to content

Latest commit

 

History

History
97 lines (72 loc) · 3.26 KB

README.md

File metadata and controls

97 lines (72 loc) · 3.26 KB
Raw

GPU version of KinectFusion with CUDA

Requirements for GPU version of KinectFusion with CUDA

We implement KinectFusion using C++ and CUDA (specifically CUDA 11) to meet the real-time computing requirements. Essential libraries such as OpenCV for image processing and Eigen for efficient matrix and vector operations will also be utilized.

Regarding the dataset, we will use the TUM RGB-D Dataset, to validate our reconstruction implementation.

Getting Started

Prerequisites

  • C++17
  • CUDA 11
  • OpenCV 4
  • Eigen 3

Install CUDA 11

Please refer to the NVIDIA website.

Install OpenCV 4 with CUDA support

This is a guide to installing OpenCV 4 with CUDA support. Here it is assumed that you have already installed CUDA 11 and downloaded the OpenCV source code to ~/Libs/opencv/4.8.0 and the OpenCV contrib source code to ~/Libs/opencv/contrib-4.8.0. Please refer to the official OpenCV installation guide.

# change the directory to the directory where you downloaded the OpenCV source code
cd ~/Libs/opencv/4.8.0 
mkdir build && cd build
# change the install directory to your preferred directory and set the path to the OpenCV contrib source code
cmake \
-D CMAKE_BUILD_TYPE=RELEASE \
-D CMAKE_INSTALL_PREFIX=~/Libs/opencv/4.8.0/install \
-D OPENCV_EXTRA_MODULES_PATH=~/Libs/opencv/contrib-4.8.0/modules \
-D WITH_CUDA=ON \
-D BUILD_DOCS=ON \
-D BUILD_EXAMPLES=OFF \
-D BUILD_TESTS=OFF \
-D BUILD_PERF_TESTS=OFF \
..
make -j16
make install

Download the TUM RGB-D Dataset

# make sure you are in the *root directory* of this project
cd KinectFusion
mkdir Data && cd Data
wget https://vision.in.tum.de/rgbd/dataset/freiburg1/rgbd_dataset_freiburg1_xyz.tgz
tar -xvzf rgbd_dataset_freiburg1_xyz.tgz

Set the path to your OpenCV and Eigen

# make sure you are in the *KinectFusion_GPU_CUDA* directory under the root directory of this project
cd KinectFusion_GPU_CUDA

# create a new cmake configuration file (or you can use any text editor to create this file)
touch config.cmake
echo "set(OpenCV_DIR /path/to/your/opencv)" >> config.cmake # set the path to your OpenCV 4
echo "set(Eigen3_DIR /path/to/your/eigen)" >> config.cmake # set the path to your Eigen 3

# set your GPU Compute Capability, refer to https://developer.nvidia.com/cuda-gpus
# for example, if you have a GTX 1080, the compute capability is 6.1. You can set it as follows:
echo "set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS};-O3;-std=c++11 -gencode arch=compute_61,code=sm_61 --expt-relaxed-constexpr)" >> config.cmake

For example, you can set the path to your OpenCV as follows:

# config.cmake
SET(OpenCV_DIR "~/Libs/opencv/4.8.0/install/lib/cmake/opencv4")

Building

# make sure you are in the *KinectFusion_GPU_CUDA* directory under the root directory of this project
cd KinectFusion_GPU_CUDA
mkdir build && cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
make -j16

Running

cd KinectFusion_GPU_CUDA/build
./KinectFusion_CUDA

Results

The reconstructed 3D model and output videos will be stored in the output directory.