Stuy Vision 2016

A library for interactive computer vision development, forked from a project by two StuyPulse alumni.

Features

• Generically read frames from cameras, videos, and images with CaptureSource
• Automatic resizing of capture source frames to desired size
• Map CaptureSources to VisionModules to compare how algorithms work on various capture sources

And, while you're using the GUI:

• Display images to the GUI at intermediate points during processing, to see what each step in the algorithm is doing
• Display arbitrary text ("tags") under images posted by vision modules
• Dynamically tune vision module variables through the GUI with:
  • automatically generated sliders for integer and floating point variables declared in vision modules
  • automatically generated checkboxes for boolean variables declared in vision modules

Installing OpenCV on your machine

If JAVA_HOME is not set, set it with

$ export JAVA_HOME=/path/to/jdk

This should be in your ~/.bashrc or .profile to persist across reboots. (Do not include the dollar sign.)

To install OpenCV 3.0, first update your packages:

$ sudo apt-get update
$ sudo apt-get upgrade

, then install OpenCV's dependencies. Running install-opencv-unix.sh will show you the dependencies, then ask you whether to continue. Say no and install them.

Then run:

$ ./install-opencv-unix.sh

Building stuyvision-lib with Ant

Make sure you have Apache Ant installed. You can check by running ant -version.

Run ant dist, or just

$ ant

to build the project and create dist/stuyvision.jar.

If javac canot find the javafx package, you are likely using an old (pre version 8) version of the JDK. If JDK version 8 is not installed, install it. It may already be installed but not used by default, in which case run:

$ sudo update-alternatives --config java
$ sudo update-alternatives --config javac

and select Java 8.

If you for some reason cannot update to a JDK that has Java 8, get the JavaFX runtime from lib/ext/jfxrt.jar in the JRE folder of a machine that does have Java 8, and put it in lib/jfxrt.jar here.

Configuring camera settings

setup-camera.sh configures settings like exposure and brightness. It configures the camera at /dev/video0 by default, but you can configure a different one with with ./setup-camera.sh -d N (to use /dev/videoN) or with ./setup-camera.sh -i (to interactively choose a device).

The script uses V4L (Video4Linux) to set the settings. V4L should be installed if you followed the OpenCV installation directions above.

If you want to install V4L on the roboRIO, use opkg:

$ opkg update
$ opkg install v4l-utils

Setting up CV on a NVIDIA Jetson ("Tegra")

To install OpenCV 3.0, run:

$ ./install-opencv-jetson.sh

Again, make sure to install the dependencies.

Running your code on startup

To run your code automatically when the Tegra boots, you'll set up a runlevel. Most Linux machines have 7 runlevels, numbered 0 through 6, in which 0 shuts down the system, 6 reboots, and 1 through 5 startup the machine in various different ways. You can read more about runlevels here and runlevels in Debian in particular here.

We will show setup of runlevel 4, a runlevel open for customization in Debian and LSB (Linux Standard Base)-compliant distributions.

What will run in runlevel 4 is determined by the contents of /etc/rc4.d/. This directory contains symbolic links to the scripts that will run when booting to this run level.

Add a symbolic link to a script which will run your code. Give it a name beginning with S, followed by a two-digit number, and then a descriptive name. The two-digit number determines the order in which the scripts will run.

E.g.:

$ vim /path/to/script # Write all that scripty goodness
$ cd /etc/rc4.d/
$ sudo ln -s /path/to/script S80run-cv

An example startup script can be seen here.

In order to set runlevel 4 as the default runlevel, open /etc/init/rc-sysinit.conf

Find the line env DEFAULT_RUNLEVEL=2, and change it to

env DEFAULT_RUNLEVEL=4

Other CV resources

• Team694/cv-edu-2017 is an example of a project using this library for simple vision processing.

• Our members have given a number of talks on CV. The slides are public: a high-level talk, a more detailed one with some code, and a detailed walkthrough with lots of code. Also, Team 254 gave a presentation on integrating CV with motion control.

• Official OpenCV 3.1.0 JavaDocs.

• Official OpenCV 2.4 documentation. Although it's not the latest version, these docs are filled with interesting and approachable examples.