xiaozhuchacha / Kinect2Toolbox

Code for CVPR 2015 Paper Understanding Tools: Task-Oriented Object Modeling, Learning and Recognition.

Home Page:https://yzhu.io/publication/tool2015cvpr/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Kinect2 Toolbox

Code for CVPR 2015 Paper

Understanding Tools: Task-Oriented Object Modeling, Learning and Recognition

Project website: http://www.yzhu.io/projects/cvpr15_tool/

Bibtex:

@InProceedings{zhu2015tool,
    title={Understanding Tools: Task-Oriented Object Modeling, Learning and Recognition},
    author={Zhu, Yixin and Zhao, Yibiao and Zhu, Song-Chun},
    booktitle={IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
    year={2015}
}

Kinect v2 Toolbox Manual

Contact: Yixin Zhu (yixin.zhu@ucla.edu)

Last revised: October 2016

Prerequisite

  • Kinect SDK Download Link.
  • Windows 8.1/10.
  • Visual Studio 2015 installed or Runtime Libraries installed.
  • MATLAB R2014b or newer version for visualization (optional).

Usage

The toolbox includes two types of code: Recorders and Converters. Recorders dump the Kinect raw data into RGB images, depth images and skeleton data. Converters post-process the data collected by Recorders.

Recorders include two sets:

Set 1: Everything will be converted to depth space later.

  • ColorRecorder: dump raw RGB images to hard drive
  • DepthRecorder: dump raw depth images to hard drive
  • ColorDepthRecorder: dump both raw depth and RGB images to hard drive
  • SkeletonRecorder: dump skeleton data to hard drive
  • ColorDepthSkeletonRecorder: dump RGB, depth and skeleton simultaneously to hard drive. Two versions are provided. The Mask version marks the skeleton area in the depth image. The mapper produced in this code will be used to convert rgb to depth space..

Set 2: Everything will be converted to color space later.

  • ColorDepthSkeletonRecorder2: dump RGB, depth and skeleton simultaneously to hard drive. The mapper produced in this code will be used to convert depth to rgb space..

Converters include two sets:

Set 1: Everything will be converted to depth space later.

  • RGBDAlign: align RGB images to the depth images
  • Depth2PC: convert raw depth images to point cloud .ply files (no color)
  • ColorDepth2PC: convert aligned RGB-D images to point cloud .ply files

Set 2: Everything will be converted to color space later.

  • RGBDAlign2: align depth images to the RGB images

We recommend to use the binary files located inside Pipeline forder if you do not need to change anything in the source code.

Record Kinect Raw Data with Kinect Studio

You will need Kinect v2 to physically connect to your PC. Please refer to this page to check whether your machine is compatible with Kinect v2. The most important component is a USB 3.0 port.

Open Kinect Studio, and switch to Record tab. Click on Connect button, and check all 11 streams. Once you finish recording, a .xef file will be generated.

alt text

Kinect v2 consumes lots of memory. If you need to record long clips, remember to edit the buffer size before recording. Also, in order to achieve high frame rate when dumping the data into hard drive, it is recommended to have more than 16GB memory for Kinect Studio.

alt text

Recorders

In this section, NO Kinect v2 needed to physically connect to PC. Following instructions assume no connection.

  • Load the recorded Kinect data (.xef file) that you want to dump using Kinect Studio.
  • In PLAY tab, click Connected button.
  • Check all 11 streams.

alt text

  • Hit Play button, and you should be able to see the replayed videos. You can change the stream (RGB, depth, skeleton, etc) in two viewers by clicking on the gear-shape buttons.

alt text

  • Hit Pause button at the beginning of the data that you want to dump.
  • Open any Recorder mentioned in previous section. You should see a stilled skeleton in a pop-up windows. In this manual, we will use ColorDepthSkeletonRecorder as the example, since it is the most comprehensive Recorder that records RGB, depth and skeleton data simultaneously.

alt text

  • Hit Play button to resume playing. The skeleton in the pop-up windows should start to move together with the videos in Kinect Studio. Data starts to dump into the datafolder.
  • Hit Pause button again to stop dumping (before the video runs out), and close the pop-up window.

Converters

All the converters load the data from data folder, and write the new data into the same data folder.

The inputs and outputs of converters are defined as

Converters Inputs Outputs Physically plugin Kinect?
Depth2PC raw_depth point cloud file (.ply) Yes
RGBDAlign depth2rgb_mapper, raw_depth, raw_rgb aligned_rgb No
ColorDepth2PC aligned_rgb, raw_depth point cloud file (.ply) Yes
RGBDAlign2 depth2rgb_mapper2, raw_depth, raw_rgb aligned_depth No

MATLAB Visualization (optional)

This part of toolbox is designed for visualization only. You can visualize the data easily in other programming languages, e.g. Python, C++, etc.

  • Copy all the files from previous data folder into the subfolder data located in the folder matlab_viz.
  • Run viz_result.m. It should show point cloud with human skeleton. The function showPointCloud is only supported in MATLAB 2014b or newer version, but can be easily replaced by function plot3 in older versions.

alt text

About

Code for CVPR 2015 Paper Understanding Tools: Task-Oriented Object Modeling, Learning and Recognition.

https://yzhu.io/publication/tool2015cvpr/

License:MIT License


Languages

Language:C++ 88.7%Language:MATLAB 9.4%Language:C 1.9%