This is a playground project to try out some features of the OpenNI2 project along with the primesense python wrapper on a Raspberry Pi 3 Model B+. As a sensor I am using the ASUS Xtion PRO LIVE.

• Raspberry Pi 3 Model B+
• ASUS Xtion PRO LIVE RGB and depth sensor

As a first step I wanted the Raspberry Pi (with Raspbian installed), OpenNI2 and the ASUS sensor to be working together successfully. To test that I used the sample program SimpleRead which comes with the OpenNI2 project (inside the Samples subfolder). To be able to execute it I had to build and install the OpenNI2 SDK on my Raspberry Pi.

First I cloned the OpenNI2 project's develop branch onto my Raspberry Pi.

git clone --single-branch -b develop https://github.com/occipital/OpenNI2.git

Then I installed the dependencies listed in the prerequisites section of the OpenNI2 README.

Next I had to make some adjustments to be able to compile and afterwards install OpenNI2 successfully. The sources that helped me to figure it out are listed below.

• OpenNI2 README itself, especially the build prerequisites when building for Linux and the packaging part.
• Getting Raspberry Pi, OpenNI2, and Asus Xtion Pro Live To Work Together
• Depth Imaging with the Asus Xtion Pro Live Sensor and the KIPR Link (Part 1)
• OpenNI2.2 ARM Binaries on Raspberry Pi
• Problem when run 'ReleaseVersion.py': subprocess.CalledProcessError
• Issue running ReleaseVersion.py
• GCC 7.2 Array subscript is below array bounds

• in file /OpenNI2/ThirdParty/PSCommon/BuildSystem/Platform.Arm
  • I commented out line 4 CFLAGS += -march=armv7-a -mtune=cortex-a9 -mfpu=neon -mfloat-abi=softfp #-mcpu=cortex-a8 and instead added CFLAGS += -mtune=arm1176jzf-s -mfpu=vfp -mfloat-abi=hard
  • commented out line 12 DEFINES += XN_NEON
• in file /OpenNI2/Packaging/ReleaseVersion.py
  • in lines 198 and 202 I changed the make calls by replacing the part where it says '-j' + calc_jobs_number() by just -j1
  • furthermore I passed another parameter ALLOW_WARNINGS=1 to the make calls

After those changes in the sources I was able to compile by navigating inside the folder /OpenNI2/Packaging and then performing the command sudo python2 ReleaseVersion.py Arm.

The results of the build can be found inside the folder OpenNI2/Packaging/Final.

I extracted the installer archive inside the folder /usr/local/src. By navigating inside the extracted folder and calling sudo chmod +x install.sh and sudo ./install.sh I was able to install the OpenNI2 SDK.

Next I tried out the example program SimpleRead by navigating inside the folder Samples/Bin and then calling sudo ./SimpleRead. The result was a continuous stream of sensor data similar to one of the resources I used.

Next I tested some sample projects using the primesense python wrapper and OpenCV to get some visual results. The original examples can be found here and are only customized a little bit.

To be able to run the examples I followed this guide to install OpenCV on my Raspberry Pi, first. Inside the virtual environment cv (created during the guide) I also installed the primesense package using the command pip install primesense.

The different stream examples can be found inside the examples subfolder. To run one example just run main.py and pass one of the available stream types (rgb, depth or ir) as a positional argument (e.g. python main.py rgb).