Raspberry Pi Setup

by Chris Lovett and Ofer Dekel

Most Embedded Learning Library (ELL) tutorials follow a common workflow. You typically start by designing an ELL model on a laptop or desktop computer. Then, you transition to your Raspberry Pi device to compile and run your application. This means that you’ll need to set up your computer and your Pi appropriately before starting any of the ELL tutorials.

Find setup instructions for your computer here:

This tutorial provides step-by-step instructions for setting up your Raspberry Pi device.

Adapters, cables, and software for your Raspberry Pi device

  • Power adapter and power cable. AI workloads use a lot of power, so be sure to use a high quality power supply. If you use a USB power adapter and micro-USB cable, choose an adapter rated for at least 12 watts (2.4 Amps) per USB port. Suggested models include 12-watt USB power adapters from Apple, Anker, and Amazon Basics.

Avoid long and thin micro-USB cables because using them will often create a noticeable voltage drop and they may fail to provide sufficient power to the Raspberry Pi device.

  • Operating system. These tutorials prefer that the operating system running on your Pi is Raspbian Stretch so if your Pi is running Jessie you will need to upgrade. If you really need to use Jessie for some reason see Troubleshooting below.

  • Network. You’ll use a network to download required software to the Pi and to transfer compiled ELL models from your computer to the Pi. Connect your Pi to your network, either over Wifi or with an Ethernet cable.

  • Pi Camera By default, OpenCV can read images from a USB webcam, but not from the Raspberry Pi camera. To enable the Pi camera, first make sure that the camera interface is enabled by running the Pi configuration tool.
      sudo raspi-config
    1. Select 5 Interfacing Option and press Enter.
    2. Select P1 Camera and press Enter .
      1. Select Yes to enable the camera interface.
    3. Install the camera module, as follows.
        sudo modprobe bcm2835-v4l2
    4. If this works you should get this device should show up /dev/video0
    5. The camera will be disabled automatically after reboot (it’s a security feature), so if you want the camera to be on after reboot without having to repeat the above steps you need to add the following line to the end of your /boot/config.txt file:
        # enable the camera
    6. And the following line to the end of your /etc/modules file:
        # camera with v4l2 driver
  • Secure Shell (SSH) The tutorials require you to copy files to the Raspberry Pi device. A typical way to copy files to the Pi is to use an SSH tool, such as the Unix scp tool or the Windows WinSCP tool. To enable SSH on your Pi, run the Pi configuration tool.
      sudo raspi-config
    1. Select 5 Interfacing Options and press Enter.
    2. Select P2 SSH and press Enter.
    3. Select Yes to enable the SSH server.

Programming tools


You’ll use CMake on the Raspberry Pi to create Python modules that can be called from the tutorial code. To install CMake on your Pi, connect to the network, open a terminal window, and type the following.

sudo apt-get update
sudo apt-get install -y cmake


This is a library for fast linear algebra operations, which can significantly increase the speed of your models. It is optional, but highly recommended. To install OpenBLAS, type the following.

sudo apt-get install -y libopenblas-dev


This is a command line tool used to transfer data via URL. When files are required to be downloaded from a URL, the instructions assume you have curl available to perform the download. To install curl, type the following.

sudo apt-get install -y curl

Python 3.4 via Miniconda

All of the tutorials require Python 3.4 or above on the Pi (and Python 3.6 on your computer). An easy way to install Python and all the required modules is with Miniconda.

Note: Miniconda is not available for Raspberry Pi Zero. Turns out Rasbian Stretch already comes with Python 3.5, so we only really need Miniconda to get our build of OpenCV which we use in the tutorials. On Raspberry Pi Zero you will need to jump to these instructions on how to Build OpenCV for Python 3.5

To install Miniconda, type the following.

curl -O --location http://repo.continuum.io/miniconda/Miniconda3-latest-Linux-armv7l.sh
chmod +x Miniconda3-latest-Linux-armv7l.sh

When prompted to install, reply [yes] to add Miniconda3 to your PATH. Then, create an environment, as follows.

source ~/.bashrc
conda create --name py34 python=3.4
source activate py34

Remember to run source activate py34 each time you start a new terminal window.


OpenCV is a computer vision library that makes it easy to read images from a camera, resize them, and prepare them for processing by ELL. To install OpenCV, type the following.

conda config --add channels "microsoft-ell"
sudo apt install -y libjasper1
conda install -y -c microsoft-ell/label/stretch opencv

Test that you can now load OpenCV with the following python script:

import cv2

You should see the output version number. If this produces an error instead and you are using a Lite version of Raspbian then you may also need to install the following because they are not provided by default with Raspbian Lite:

sudo apt-get install -y libharfbuzz-dev 
sudo apt-get install -y libgtk2.0-dev libgtk-3-dev
sudo apt-get install -y libjpeg-dev libtiff5-dev libjasper-dev libpng12-dev libdc1394-22-dev
sudo apt-get install -y libavcodec-dev libavformat-dev libswscale-dev libv4l-dev
sudo apt-get install -y libxvidcore-dev libx264-dev

C++ OpenCV SDK

If you intend to run the C++ tutorials, you also need the C++ OpenCV SDK, which you can install on your Raspberry Pi using the following command.

sudo apt-get install -y libopencv-dev

Performance tips

Raspberry Pi devices weren’t designed to run AI workloads. Many AI workloads, like visual object tracking and audio keyword detection, run continuously for long periods of time and require near real-time responsiveness. You may want to disable features such as Energy Star, screensaver, and dynamic clocking in order to get the best performance for AI workloads.

Disable Energy Star and Screensaver

Edit the file ~/.config/lxsession/LXDE-pi/autostart, for example, by typing the following.

nano ~/.config/lxsession/LXDE-pi/autostart

Add the following lines to this file.

@xset -dpms
@xset s noblank
@xset s off

The first line disables Energy Star and the next two lines disable the screensaver. These changes take effect after rebooting.

Disable dynamic clocking

The Raspberry Pi supports dynamic clocking, which means that it can change the processor frequency according to processor load. You can check the range of processor frequencies that your Raspberry Pi is configured for by typing lscpu. To disable dynamic clocking, edit /boot/config.txt, for example, by typing the following.

sudo nano /boot/config.txt

Add the setting.


This change takes effect after rebooting.

Heat management

Computation produces heat. Even at the default processor frequency of 700 megahertz (MHz), a Raspberry Pi running a large AI workload at room temperature can overheat unless it is fitted with a physical cooling device. (See the tutorial about actively cooling your Pi.) If you don’t want to physically cool your Pi, you can cool it by reducing the processor frequency (underclocking).

If you do fit your Pi with an active cooling attachment, you can also increase the processor frequency (overclocking) without overheating.

Warning Overclocking your Raspberry Pi will void your warranty. Overclocking might cause your Raspberry Pi to freeze or lose USB peripheral connections. Proceed at your own risk.

To change your processor frequency, edit /boot/config.txt, for example, by typing the following.

sudo nano /boot/config.txt

The default processor speed is 700 MHz. To underclock your processor, add the setting


The change takes effect after a reboot. After the Pi reboots, you can confirm the change using the utility lscpu.

To check if a frequency of 600 MHz is slow enough, run your AI workload and measure the processor temperature. Measuring temperature of your Pi as follows.

watch /opt/vc/bin/vcgencmd measure_temp

This command prints the processor temperature every 2 seconds. When fully stressing your Pi, you want the processor temperature to stay far below 85 degrees Celsius. Try keep the processor temperature below 60 degrees Celsius.

To overclock the processor, experiment with frequencies of 800, 900, 1,000, and even 1,200. You will also want to adjust some of the other parameters, such as sdram_freq and over_voltage. Try the following configuration.



Raspbian Jessie

If you really need to use Raspbian Jessie then you will need to install a different version of OpenCV. To do that run the following:

conda install -c microsoft-ell/label/jessie opencv

ImportError: libavcodec.so.57: cannot open shared object file: No such file or directory

You might be running Rasbian Jessie but you installed OpenCV for Stretch. You can check this by running lsb_release -a. If it says ‘Codename: jessie’ then you need to upgrade to Raspbian Stretch or install Jessie version of OpenCV, see above.

ImportError: libavcodec.so.56: cannot open shared object file: No such file or directory

You might be running Rasbian Stretch but you installed OpenCV for Jessie. You can check this by running lsb_release -a. If it says ‘Codename: stretch’ then you need to install the OpenCV version for Stretch which you can do with this command:

conda install -c microsoft-ell/label/stretch opencv

ImportError: No module named ‘numpy’

You probably forgot to activate your Miniconda environment using source activate py34. See Miniconda instructions above.

Gtk-WARNING : cannot open display

You are probably trying to use OpenCV from an SSH terminal window. You must tell OpenCV which display to use with the command export DISPLAY=:0.

Note that many of the tutorial scripts terminate when they receive keyboard input - to terminate a script over SSH, type CTRL+C.