Version 38 (modified by 7 years ago) (diff) | ,
---|
Vision Framework
Mentors: Joy Cho
Students: Reid Huntley, Benji Albert, Nathan Misner
Background
From 2015-2017, team 2537 developed a machine vision platform for autonomous targeting and navigation. It was first used successfully during the 2016-2017 season and helped the team reach the finals at the world championship. The vision platform is based on the Raspberry Pi 3 with camera module in a clear case. The 2016-2017 version used an MP1584EN power supply to convert the 12v battery voltage to the 5v required for the machine vision hardware. It also included a transistor switch to allow the Pi to control power to a green 12v LED light ring. Communication between the Pi and roboRIO was uni-directional using PWM; the PWM duty cycle represented the angle for the robot to rotate to be perfectly aligned with the target (the gear peg in Steamworks). Many lessons were learned including:
- Mount the vision system securely, the more it moves relative to the robot, the less accuracy it will deliver
- Don't try to use vision while the robot is driving, the vibration makes the images move too much
- Protect your vision module from collisions: we used a custom polycarbonate shield
2017-2018 Objectives
Extend the vision module software to make it more accessible to other teams and more user friendly generally. Changes planned include:
- Execute a base program on startup that continuously reads the camera feed (subscribes to camera) and publishes corresponding data via the Pi's serial output (located on the Pi's GPIO connector). This data can then be fairly simply read by the RoboRIO.
- Improve processing speed to yield better resolution/frame rate
- Replace the 12v monochrome LED ring with a smart Neopixel-compatible 12x RGB LED light ring based on the WS2812 controller. This will allow the module to run entirely from 5v, allow teams to select their light-ring color, and even make fancy light shows. Smart light rings are available from adafruit and eBay among other sources. You can read about the neopixel here. Neopixel control can be achieved using many libraries including diozero; see also here
Materials
Vision Module
- Clear plastic case
- Raspberry Pi 3
- Pi camera module w/cable
- NeoPixel Light Ring
- Qty 3 M-F and qty 3 F-F jumper wires
- 8GB or larger Micro SD card
Additional materials needed during development
- Usb Keyboard and Mouse (for developing on the Pi)
- HDMI cable or HDMI-to-VGA adapter (for developing on the Pi)
- Monitor (that accepts HDMI or VGA input)
- Micro USB Cable
- USB charger capable of supplying 5v at 2A or greater
Raspberry Pi Setup
Raspbian Lite
Install Win32 Disk Imager
Install Raspian Lite onto MicroSD card via Win32 Disk Imager
Load Raspberry Pi 3 (rpi) with the newly imaged MicroSD
username: pi password: raspberry
connect to wifi by adding the following to the /etc/network/interfaces file:
auto wlan0 iface wlan0 inet dhcp wpa-ssid "your-ssid" wpa-psk "your-password"
reboot the pi:
sudo shutdown -r now
update with the following commands:
sudo apt-get update sudo apt-get upgrade sudo apt-get dist-upgrade sudo apt-get clean
set localization configuration options:
sudo raspi-config
reboot:
sudo shutdown -r now
X server
Install Xorg and Xinit:
sudo apt-get install --no-install-recommends xserver-xorg sudo apt-get install --no-install-recommends xinit
Install the MATE GUI:
sudo apt-get install mate-desktop-environment-core
Install LightDM login manager:
sudo apt-get install lightdm
reboot:
sudo shutdown -r now
login to MATE
At the top left:
navigate to preferences -> hardware -> keyboard shortcuts
to set Run a Terminal to Ctrl + Alt + T
or the like
open a terminal
navigate to Edit -> Profile Preferences -> Colors
to deselect Use colors from system theme
select Built-in schemas: White on black
Development Tools
Java 8
Now that the terminal color scheme is not killing you, we are going to start off by installing java:
sudo apt-get install oracle-java8-jdk
test the java installation (output should include "1.8.0_65" and should not include "openjdk"):
java -version
Let's also set the JAVA_HOME variable to be exported on startup:
echo "export JAVA_HOME=/usr/lib/jvm/jdk-8-oracle-arm32-vfp-hflt" >> ~/.bashrc source ~/.bashrc
Finally, let's install ANT:
sudo apt-get install ant
Pi4J (Java-GPIO Interface)
Now we can install pi4j, a Java interface for the pi GPIO:
curl -s get.pi4j.com | sudo bash
rpi_ws281x (NeoPixel? Control Library)
Then we can use the rpi_ws281x library for controlling the Adafruit NeoPixel lightring
But first, let's install some tools and dependencies:
sudo apt-get install build-essential python-dev git scons swig
Now we can install the rpi_ws281x library:
git clone https://github.com/jgarff/rpi_ws281x.git cd rpi_ws281x scons
OpenCV
First of all, order some pizza or something because this is going to take about 3 hours (if you are lucky).
Also, you need a heatsink for your processor before building OpenCV. If you see this overheating symbol, you will die. Straight up.
EDIT: we have tried with a heatsink and by itself, it didn't work. You need to used forced air (fan) or water *and* a heatsink to cool your pi. Please skip to the WATER COOLING section.
If you don't have a heatsink on hand, tape some pennies to the CPU like we did:
================WATER COOLING================
Poor man's water cooling system:
Make sure to change the ice cube at least once every half hour. Also, only single bag the ice cube.
Now we are going to install OpenCV and its dependencies.
We only install OpenCV for C++ and Java, not for Python (we ain't scrubs):
sudo apt-get install cmake git libgtk2.0-dev pkg-config libavcodec-dev libavformat-dev libswscale-dev cd git clone https://github.com/opencv/opencv.git cd opencv mkdir build cd build cmake .. make -j $[$(nproc)+1] sudo make install
This builds a single shared object file (libopencv_java320.so), a jar file (opencv-320.jar) and a ton of static libraries (*.a)
It copies the shared object file and the jar file to the install directory: /usr/local/share/OpenCV/java/...
NOTE: we could not get the java wrapper to work with ffmpeg enabled (we ran into a Segmentation fault with opencv libs 3.2.0, 3.3.0, and 3.3.1). We gave up and disabled it.
NOTE: All of the below steps work with OpenCV versions 3.2.0, 3.3.0, and 3.3.1. Use whichever you prefer.
If you do not want to git clone the master opencv repository, download a zipped version:
opencv-3.2.0.zip
Java Projects with OpenCV
Enable VideoCapture?
First, to allow OpenCV VideoCapture? to access the ribbon cable camera, execute the following:
echo "sudo modprobe bcm2835-v4l2" >> ~/.bashrc source ~/.bashrc
Compiling
javac -d <build_path> -classpath <external_jars> <source_path>/*.java
where:
build_path = directory to store class files
external_jars = string specifying individual jar dependencies (such as OpenCV and Pi4J) with delimiters as follows:
:/path_to_jars/jar1.jar:/path_to_jars/jar2.jar
souce_path = directory containing all of your java files to be compiled
We recommend creating a libs
folder in which you store all of your external jars such as the opencv-320.jar
and all the Pi4J jars rather than having your libraries scattered all over your machine.
Running
sudo java -Djava.library.path=<opencv_install_path> -Dpi4j.debug -Dpi4j.linking=dynamic -classpath <external_jars> <Main_class>
where:
external_jars = same string that you used in the Compilation step
Main_class = name of the class file with main method
- For example, where
Vision.class
is the class file,Main_class = Vision
- For example, where
C++ Projects with OpenCV
NOTE: our performance doubled when we transitioned to C++ from Java
With C++, we can process (capture and segment) 640x480 frames at 66 fps and 1080x720 frames at 28 fps
With Java, we could only get about 30 fps with 640x480 frames
Let's install an IDE so we do not have to keep writing in vi/vim/nano:
sudo apt-get install codeblocks
We experienced a bug that constantly crashed CodeBlocks:
To prevent crashing, navigate to: Settings -> Editor -> Code completion -> Symbols browser
and check Disable symbols browser
(the Code completion
tab is on the west panel of the configure editor.)
Navigate to: File -> New -> Project -> Console Application
And fill in the fields that come up.
If your Management
tab is not visible, open it by selecting: View -> Manager
Also, if your Logs & others
tab is not open already, select View -> Logs
Now right click your project in the Management
tab and navigate to: Build options... -> Linker settings
Under the Link libraries
panel, select Add
and select all OpenCV shared object (.so) files located in /home/pi/opencv/build/lib/*.so
(You can [ctrl + left-click] multiple *.so] files to add them all at once)
Now let's navigate to: Search directories -> Linker
In the Linker
tab, select Add
and add the path to your opencv lib: /home/pi/opencv/build/lib
Tricks and Tips
Using OpenCV with the Raspberry Pi 3 ribbon cable camera is a little cumbersome because you cannot use many of the VideoCapture?.set(char*,int) functions.
Instead, we use system commands to set camera settings via the v4l2-ctl library.
For FRC retroreflective tape segmentation, use the following system commands to prevent auto adjustment features such as auto exposure
, white balance
, and exposure time
:
- v4l2-ctl --set-ctrl=auto_exposure=1
- v4l2-ctl --set-ctrl=white_balance_auto_preset=0
- v4l2-ctl --set-ctrl=auto_exposure_bias=0
- v4l2-ctl --set-ctrl=exposure_time_absolute=100
Make your C++ program set these options by using the stdlib.h
system(char*)
command which simply takes a string command as input.
Alternatively you could add these options to your bashrc
file like we did with the modprobe, but you may want to tweak these options or experiment with them, so putting them into your C++ program makes it a little easier to manipulate.