Lab Section 1: Point Cloud Fusion

This lab section teaches you how to capture raw data and fuse all point cloud into one, with additional point cloud processing steps in Meshlab.

System Requirements

• Windows 10 64 bit
• If you are using Mac or Linux, plz contact Yingjun through yingjun.tian@postgrad.manchester.ac.uk and I would arrage your time slot for using our lab Windows computer

Software Requirements

• Python >= 3.0

  • access this website to download windows 64bit installer:

    https://www.python.org/downloads/release/python-387/

  • During installation, remember click "add to system path" option

  • When you want to run python script like "cmd.py":

    • click on main menu of windows, type in "cmd" to open windows terminal
    • change directory to your working directory like: "cd C:/workingDirectory"
    • Runn .py file, plz use "python cmd.py" or "cmd.py" to run the program

Program details

• First Lab Section includes:
  • Capturing to acquire dataset
  • Image conversion
  • Data Fusion
  • Data Cleaning

Capturing: Acquire Dataset

• You could get dataset from two ways

  • 3 datasets provided by TA
  • You could use capture program provided by us to capture rgb and depth image

• Remember the image size of your dataset

  • the image size of dataset provided by TA is included in dataset folder
  • capturing dataset: 1280X720

Image Conversion

• This bin file will convert image from png to ppm or pgm (fusion program requests this format)

• Run ./Image_Conversion/cmd.py

  • Before running, you should provide how many rgb file or depth file you have

    By changing ""png_file_number" variable value

• After conversion, you would get converted rgb and depth files in rgb_ppm and depth_pgm folder

Camera Calibration (Optional)

• Calibrate Realsense D4XX
  • See ./Calib/bin/README.me for details
  • We provide chess.pdf in ./Calib/bin/ to automatically calibrate camera and calibration program will provide output txt file which can be directly read from Fusion exe file
  • Remark: This bin file is from https://github.com/carlren/OpenNICalibTool
• If you are not interested in this process, plz ignore this part, we have provided some calculated calibration files for you to use

Data fusion

• move the converted ppm and pgm folders to **./infiniTAM_cli_bin **
• Fusion result will be stored in ./infiniTAM_cli_bin/FusionResult
• Run ./infiniTAM_cli_bin/cmd.py to get fusion result
  • Before starting, please provide image size you are using, and please see cmd.py for further detail about image size of every dataset:

Data Cleaning

You could download this opensource software through this website

Website: https://www.meshlab.net/

We choose to use MeshLab to clean the fused point cloud to be clean point cloud:

• Outlier filtering

• Other point cloud selection

• Mesh Selection and cleaning operation

Need to point out:

• Fused result is mesh, but the mesh quality is not good because the point cloud is noisy
• So we should only use the point cloud data and do some mesh reconstruction in our second lab section