Installing a PCL with KINFU and GPU parallel operation versions

After seeing the Kinect fusion effect for a few days, I was interested in it, so I wanted to build a Kinect fusion frame to play with, but the Kinfu open source code in PCL has been a beta version, and there's no release. Then can only silently uninstall the previously installed version 1.7.1 to replace the PCL test version.
The PCL beta version can also be downloaded on the GitHub home page, just select the master branch of the home page and download into zip, and the Kinfu code is in the GPU folder.

Compared to installing version 1.7.1, you want to run Kinfu, you need to speed up the GPU with the graphics card, the third-party libraries that need to be installed in advance, and a copy of all of my third-party libraries:
　　

A. Specific installation steps:

• Prepare beforehand:

  1. Confirm that your computer graphics card is n cards, and the computing power is above 2.x (Fermi) or 3.x (Kepler);
  2. Go to NVIDIA official website Download Cuda latest version, and update own graphics driver, my cuda is 6.5 version, video card is 650M;
  3. or update your graphics driver, the newer driver of the graphics card will make the final Kinfu frame rate faster!
  4. Install third-party libraries, BOOST EIGEN FLANN qhull VTK qt4.8.0, clean up the original third-party libraries before installation;

• To start the installation:

  1. CMake compiled source code, configure will appear after the first time a lot of red build, tick Build_gpu can be, online also said tick Build_tools build_visulation Build_cuda Build_apps, But my first two here is the default check, Build_cuda Build_apps I finally did not choose;
     
  2. Before clicking on generate, take a closer look at the output below CMake, to built contains all the functions you need;
  3. Click Generate;
  4. VS Open the generated function to compile and install;

Two. Tips and Bugs:

2. Running KINFU requires the support of Openni and PrimeSense, and the PCL captures point clouds from the Kinect is based on the Openni framework, and it is important to note that the Microsoft Official SDK for Kinect must not be installed in advance because the official will block the PCL Openni interface, The final PCL does not read the depth data, showing

   Can ' t open depth source

4. When testing, you can first generate the project Openni-viewer, test the PCL can successfully obtain the Kinect data, and then generate Kinfu-app to test, the final error after generating all the PCL, and install;

6. This version of the PCL source Problem 1:
   When compiling in the source code, the BUILD_GPU will appear regardless of the selection

   Libboost_thread-vc100-mt-gd-1_50.lib (thread.obj): Error LNK2019

   This error.
   Article http://blog.csdn.net/lming_08/article/details/19114417
   Give the solution to this problem, in the source file Cmake/pcl_find_boost.cmake

   Required boost Modules Find_package (boost 1.40.0 Required components
   System filesystem thread Date_time IOStreams)

   Revision changed to

   Required boost Modules Find_package (boost 1.40.0 Required components
   System filesystem thread Date_time IOStreams chrono)

8. This version of the PCL source Problem 2:
   Would have thought this is going to be done, did not expect 1 hours after the compilation result is 130+ success, 2 failed, then particularly frustrated. The reason for the failure is that there was a PNG error, and the map is as follows:
   
   The solution that is open on the net is to block out the cmakelist after opening in the Source Tools folder.

   Pcl_add_executable (pcl_pcd2png "${subsys_name}" pcd2png.cpp)
   Target_link_libraries (pcl_pcd2png Pcl_common pcl_io)
   Pcl_add_executable (pcl_organized_pcd_to_png "${subsys_name}" Organized_pcd_to_png.cpp)
   Target_link_libraries (pcl_organized_pcd_to_png Pcl_common pcl_io)

   These four sentences can be used, and eventually the complete PCL can be installed correctly

10. Another problem with the installation process is that after uninstalling the Kinect's Microsoft Official SDK, you uninstall the PCL Openni and Premesense and reinstall it, but after installation, CMake will not be able to find Openni, KINFU and GPU engineering, Can not find the specific reason, the next day first use CMake compile pcl1.7.1 source to find Openni, then go to the new Pcl-master source containing Kinfu, at this time only successfully found Openni, and finally successfully installed, the cause of the problem did not delve into later.

The following a few kinfu, a simple test, RGB images are collected by time interval, so some places have models but no texture, and the texture of the intersection of the place does not do texture fusion, these two issues are I will be ready to optimize the direction of the code.

Add: I downloaded the pcl-master source code, is in the November 8, 2014 version, the official follow-up version should be fixed in succession to the bug.

　　　　　　　

(reprint please indicate author and source: http://blog.csdn.net/gh234505 do not use for commercial use without permission)

Installing a PCL with KINFU and GPU parallel operation versions