stbt

A video-capture record/playback system for automated testing of set-top boxes

Copyright

Copyright (C) 2012-2013 YouView TV Ltd. and others

License

LGPL v2.1 or (at your option) any later version (see LICENSE file in the source distribution for details)

Version

@VERSION@

Manual section

1

Manual group

stb-tester

SYNOPSIS

stbt record [options]

stbt run [options] [script]

DESCRIPTION

stbt record will record a test case by listening for remote-control keypresses, taking screenshots from the set-top box as it goes.

You then (manually) crop the screenshots to the region of interest.

(Optionally) you manually edit the generated test script, which will look something like this:

press("MENU")
wait_for_match("Guide.png")
press("OK")
wait_for_match("BBC One.png")

stbt run will play back the given test script, returning an exit status of success or failure for easy integration with your existing test reporting system.

stbt has other auxiliary sub-commands; run stbt --help for details.

OPTIONS

Global options

--control=<uri>

A remote control to use for controlling the set top box. uri can be:

lirc:([<lircd_socket>]|[<hostname>:]<port>):<remote_control_name>

A hardware infrared emitter controlled by the lirc (Linux Infrared Remote Control) daemon.

• If lircd_socket is specified (or none of lircd_socket, hostname and port are specified) remote control commands are sent via a lircd socket file. lircd_socket defaults to /var/run/lirc/lircd.
• If port is specified, remote control commands are sent via a lircd TCP listener on localhost.
• If hostname and port are specified, remote control commands are sent via a lircd TCP listener on a remote host.

remote_control_name is the name of a remote-control specification in lircd.conf.

Examples:

lirc::myremote
lirc:/var/run/lirc/lircd:myremote
lirc:8700:myremote
lirc:192.168.100.100:8700:myremote

irnetbox:<hostname>:<output>:<config_file>

RedRat irNetBox network-controlled infrared emitter hardware. hostname is the hostname or IP address of the irNetBox device. output is the infrared output to use, a number between 1 and 16 (inclusive). config_file is the configuration file that describes the infrared protocol to use; it can be created with RedRat's (Windows-only) "IR Signal Database Utility". stbt supports the irNetBox models II and III.

vr:<hostname>:<port>

A "virtual remote" that communicates with the set-top box over TCP. Requires a virtual remote listener (which we haven't released yet) running on the stb.

none

Ignores key press commands.

test

Used by the selftests to change the input video stream. Only works with --source-pipeline=videotestsrc. A script like press("18") will change videotestsrc's pattern property (see gst-inspect videotestsrc).

--source-pipeline=<pipeline>

A gstreamer pipeline providing a video stream to use as video output from the set-top box under test. For the Hauppauge HD PVR use:

v4l2src device=/dev/video0 ! mpegtsdemux ! video/x-h264 ! decodebin2

--sink-pipeline=<pipeline>

A gstreamer pipeline to use for video output, like xvimagesink.

-v, --verbose

Enable debug output.

With stbt run, specify -v twice to dump intermediate images from the image processing algorithm to the ./stbt-debug directory. Note that this will dump a lot of files -- several images per frame processed. This is intended for debugging the image processing algorithm; it isn't intended for end users.

Additional options to stbt run

--save-video=<file>

Record a video (in the HTML5-compatible WebM format) to the specified file.

Additional options to stbt record

--control-recorder=<uri>

The source of remote control presses. uri can be:

lirc:([<lircd_socket>]|[<hostname>:]<port>):<remote_control_name>

A hardware infrared receiver controlled by the lirc (Linux Infrared Remote Control) daemon. Parameters are as for --control.

vr:<hostname>:<port>

Listens on the socket <hostname>:<port> for a connection and reads a "virtual remote" stream (which we haven't documented yet, but we'll probably change it soon to be compatible with LIRC's protocol).

file://<filename>

Reads remote control keypresses from a newline-separated list of key names. For example, file:///dev/stdin to use the keyboard as the remote control input.

-o <filename>, --output-filename=<filename>

The file to write the generated test script to.

CONFIGURATION

All parameters that can be passed to the stbt tools can also be specified in configuration files. Configuration is searched for in the following files (with later files taking precedence):

1. /etc/stbt/stbt.conf
2. ~/.config/stbt/stbt.conf
3. $STBT_CONFIG_FILE

These files are simple ini files with the form:

[global]
source_pipeline = videotestsrc
sink_pipeline = xvimagesink sync=false
control = None
verbose = 0
[run]
save_video = video.webm
script = test.py
[record]
output_file = test.py
control_recorder = file:///dev/stdin

Each key corresponds to a command line option with hyphens replaced with underscores.

EXIT STATUS

0 on success; 1 on test script failure; 2 on any other error.

Test scripts indicate failure (the system under test didn't behave as expected) by raising an instance of stbt.UITestFailure (or a subclass thereof). Any other exception is considered a test error (a logic error in the test script, an error in the system under test's environment, or an error in the test framework itself).

HARDWARE REQUIREMENTS

The test rig consists of a Linux server, with:

• A video-capture card (for capturing the output from the system under test)
• An infrared receiver (for recording test cases)
• An infrared emitter (for controlling the system under test)

Video capture card

You'll need a capture card with drivers supporting the V4L2 API (Video-for-Linux 2). We recommend a capture card with mature open-source drivers, preferably drivers already present in recent versions of the Linux kernel.

The Hauppauge HD PVR works well (and works out of the box on recent versions of Fedora), though it doesn't support 1080p. If you need an HDCP stripper, try the HD Fury III.

Infra-red emitter and receiver

An IR emitter+receiver such as the RedRat3, plus a LIRC configuration file with the key codes for your set-top box's remote control.

Using software components instead

If you don't mind instrumenting the system under test, you don't even need the above hardware components.

stb-tester uses gstreamer, an open source multimedia framework. Instead of a video-capture card you can use any gstreamer video-source element. For example:

• If you run tests against a VM running the set-top box software instead of a physical set-top box, you could use the ximagesrc gstreamer element to capture video from the VM's X Window.
• If your set-top box uses DirectFB, you could install the DirectFBSource gstreamer element (https://bugzilla.gnome.org/show_bug.cgi?id=685877) on the set-top box to stream video to a updsrc gstreamer element on the test rig.

Instead of a hardware infra-red receiver + emitter, you can use a software equivalent (for example a server running on the set-top box that listens on a TCP socket instead of listening for infra-red signals, and your own application for emulating remote-control keypresses). Using a software remote control avoids all issues of IR interference in rigs testing multiple set-top boxes at once.

Linux server

An 8-core machine will be able to drive 4 set-top boxes simultaneously with at least 1 frame per second per set-top box.

SOFTWARE REQUIREMENTS

• A Unixy operating system (we have only tested on Linux and Mac OS X).
• Drivers for any required hardware components.
• gstreamer 0.10 (multimedia framework) + gst-plugins-base + gst-plugins-good.
• python (we have tested with 2.6 and 2.7; on <2.7 you will also need to install the python-argparse package) + pygst + docutils (for building the documentation) + nose (for the self-tests).
• OpenCV (image processing library) version >= 2.0.0, and the OpenCV python bindings.
• For the Hauppauge video capture device you'll need the gstreamer-ffmpeg package (e.g. from the rpmfusion-free repository) for H.264 decoding.

INSTALLING FROM SOURCE

Run "make install" from the stb-tester source directory.

See http://stb-tester.com/getting-started.html for the required dependencies and configuration.

TEST SCRIPT FORMAT

The test scripts produced and run by stbt record and stbt run, respectively, are actually python scripts, so you can use the full power of python. Don't get too carried away, though; aim for simplicity, readability, and maintainability.

The following functions are available:

press(key)

Send the specified key-press to the system under test.

The mechanism used to send the key-press depends on what you've configured with --control.

key is a string. The allowed values depend on the control you're using: If that's lirc, then key is a key name from your lirc config file.

wait_for_match(image, timeout_secs=10, consecutive_matches=1, noise_threshold=None, match_parameters=None)

Search for image in the source video stream.

Returns MatchResult when image is found. Raises MatchTimeout if no match is found after timeout_secs seconds.

consecutive_matches forces this function to wait for several consecutive frames with a match found at the same x,y position. Increase consecutive_matches to avoid false positives due to noise.

The templatematch parameter noise_threshold is marked for deprecation but appears in the args for backward compatibility with positional argument syntax. It will be removed in a future release; please use match_parameters.confirm_threshold instead.

Specify match_parameters to customise the image matching algorithm. See the documentation for MatchParameters for details.

press_until_match(key, image, interval_secs=3, noise_threshold=None, max_presses=10, match_parameters=None)

Calls press as many times as necessary to find the specified image.

Returns MatchResult when image is found. Raises MatchTimeout if no match is found after max_presses times.

interval_secs is the number of seconds to wait for a match before pressing again.

The templatematch parameter noise_threshold is marked for deprecation but appears in the args for backward compatibility with positional argument syntax. It will be removed in a future release; please use match_parameters.confirm_threshold instead.

Specify match_parameters to customise the image matching algorithm. See the documentation for MatchParameters for details.

wait_for_motion(timeout_secs=10, consecutive_frames='10/20', noise_threshold=0.84, mask=None)

Search for motion in the source video stream.

Returns MotionResult when motion is detected. Raises MotionTimeout if no motion is detected after timeout_secs seconds.

Considers the video stream to have motion if there were diferences between the specified number of consecutive_frames, which can be:

• a positive integer value, or
• a string in the form "x/y", where x is the number of frames with motion detected out of a sliding window of y frames.

Increase noise_threshold to avoid false negatives, at the risk of increasing false positives (a value of 0.0 will never report motion). This is particularly useful with noisy analogue video sources.

mask is a black and white image that specifies which part of the image to search for motion. White pixels select the area to search; black pixels the area to ignore.

detect_match(image, timeout_secs=10, noise_threshold=None, match_parameters=None)

Generator that yields a sequence of one MatchResult for each frame processed from the source video stream.

Returns after timeout_secs seconds. (Note that the caller can also choose to stop iterating over this function's results at any time.)

The templatematch parameter noise_threshold is marked for deprecation but appears in the args for backward compatibility with positional argument syntax. It will be removed in a future release; please use match_parameters.confirm_threshold intead.

Specify match_parameters to customise the image matching algorithm. See the documentation for MatchParameters for details.

detect_motion(timeout_secs=10, noise_threshold=0.84, mask=None)

Generator that yields a sequence of one MotionResult for each frame processed from the source video stream.

Returns after timeout_secs seconds. (Note that the caller can also choose to stop iterating over this function's results at any time.)

noise_threshold is a parameter used by the motiondetect algorithm. Increase noise_threshold to avoid false negatives, at the risk of increasing false positives (a value of 0.0 will never report motion). This is particularly useful with noisy analogue video sources.

mask is a black and white image that specifies which part of the image to search for motion. White pixels select the area to search; black pixels the area to ignore.

frames(timeout_secs=None)

Generator that yields frames captured from the GStreamer pipeline.

"timeout_secs" is in seconds elapsed, from the method call. Note that you can also simply stop iterating over the sequence yielded by this method.

Returns an (image, timestamp) tuple for every frame captured, where "image" is in OpenCV format.

save_frame(image, filename)

Saves an OpenCV image to the specified file.

Takes an image obtained from get_frame or from the screenshot property of MatchTimeout or MotionTimeout.

get_frame()

Returns an OpenCV image of the current video frame.

get_config(section, key, default=None)

Read the value of key from section of the stbt config file.

See 'CONFIGURATION' in the stbt(1) man page for the config file search path.

Raises ConfigurationError if the specified section or key is not found, unless default is specified (in which case default is returned).

debug(msg)

Print the given string to stderr if stbt run --verbose was given.

class MatchParameters

Parameters to customise the image processing algorithm used by wait_for_match, detect_match, and press_until_match.

You can change the default values for these parameters by setting a key (with the same name as the corresponding python parameter) in the [match] section of your stbt.conf configuration file.

match_method (str) default: sqdiff-normed

The method that is used by the OpenCV cvMatchTemplate algorithm to find likely locations of the "template" image within the larger source image.

Allowed values are "sqdiff-normed", "ccorr-normed", and "ccoeff-normed". For the meaning of these parameters, see the OpenCV cvMatchTemplate reference documentation and tutorial:

• http://docs.opencv.org/modules/imgproc/doc/object_detection.html
• http://docs.opencv.org/doc/tutorials/imgproc/histograms/template_matching/template_matching.html

match_threshold (float) default: 0.80

How strong a result from cvMatchTemplate must be, to be considered a match. A value of 0 will mean that anything is considered to match, whilst a value of 1 means that the match has to be pixel perfect. (In practice, a value of 1 is useless because of the way cvMatchTemplate works, and due to limitations in the storage of floating point numbers in binary.)

confirm_method (str) default: absdiff

The result of the previous cvMatchTemplate algorithm often gives false positives (it reports a "match" for an image that shouldn't match). confirm_method specifies an algorithm to be run just on the region of the source image that cvMatchTemplate identified as a match, to confirm or deny the match.

The allowed values are:

"none"

Do not confirm the match. Assume that the potential match found is correct.

"absdiff" (absolute difference)

The absolute difference between template and source Region of Interest (ROI) is calculated; thresholded and eroded to account for potential noise; and if any white pixels remain then the match is deemed false.

"normed-absdiff" (normalized absolute difference)

As with absdiff but both template and ROI are normalized before the absolute difference is calculated. This has the effect of exaggerating small differences between images with similar, small ranges of pixel brightnesses (luminance).

This method is more accurate than absdiff at reporting true and false matches when there is noise involved, particularly aliased text. However it will, in general, require a greater confirm_threshold than the equivalent match with absdiff.

When matching solid regions of colour, particularly where there are regions of either black or white, absdiff is better than normed-absdiff because it does not alter the luminance range, which can lead to false matches. For example, an image which is half white and half grey, once normalised, will match a similar image which is half white and half black because the grey becomes normalised to black so that the maximum luminance range of [0..255] is occupied. However, if the images are dissimilar enough in luminance, they will have failed to match the cvMatchTemplate algorithm and won't have reached the "confirm" stage.

confirm_threshold (float) default: 0.16

Increase this value to avoid false negatives, at the risk of increasing false positives (a value of 1.0 will report a match every time).

erode_passes (int) default: 1

The number of erode steps in the absdiff and normed-absdiff confirm algorithms. Increasing the number of erode steps makes your test less sensitive to noise and small variances, at the cost of being more likely to report a false positive.

Please let us know if you are having trouble with image matches so that we can further improve the matching algorithm.

class MatchResult

• `timestamp`: Video stream timestamp.
• `match`: Boolean result.
• `position`: Position of the match.
• `first_pass_result`: Value between 0 (poor) and 1.0 (excellent match) from the first pass of the two-pass templatematch algorithm.

class Position

• x and `y`: Integer coordinates from the top left corner of the video frame.

class MotionResult

• `timestamp`: Video stream timestamp.
• `motion`: Boolean result.

class MatchTimeout(UITestFailure)

• `screenshot`: An OpenCV image from the source video when the search for the expected image timed out.
• `expected`: Filename of the image that was being searched for.
• `timeout_secs`: Number of seconds that the image was searched for.

class MotionTimeout(UITestFailure)

• `screenshot`: An OpenCV image from the source video when the search for motion timed out.
• `mask`: Filename of the mask that was used (see wait_for_motion).
• `timeout_secs`: Number of seconds that motion was searched for.

class NoVideo(UITestFailure)

No video available from the source pipeline.

class UITestFailure(Exception)

The test failed because the system under test didn't behave as expected.

class UITestError(Exception)

The test script had an unrecoverable error.

TEST SCRIPT BEST PRACTICES

• When cropping images to be matched by a test case, you must select a region that will not be present when the test case fails, and that does not contain any elements that might be absent when the test case succeeds. For example, you must not include any part of a live TV stream (which will be different each time the test case is run), nor translucent menu overlays with live TV showing through.
• Crop template images as tightly as possible. For example if you're looking for a button, don't include the background outside of the button. (This is particularly important if your system-under-test is still under development and minor aesthetic changes to the UI are common.)
• Always follow a press with a wait_for_match -- don't assume that the press worked.
• Use press_until_match instead of assuming that the position of a menu item will never change within that menu.
• Use the timeout_secs parameter of wait_for_match and wait_for_motion instead of using time.sleep.
• Rename the template images captured by stbt record to a name that explains the contents of the image.
• Extract common navigation patterns into separate python functions. It is useful to start each test script by calling a function that brings the system-under-test to a known state.

SEE ALSO

• http://stb-tester.com/
• http://github.com/drothlis/stb-tester

AUTHORS

• Will Manley <will@williammanley.net>
• David Rothlisberger <david@rothlis.net>
• Hubert Lacote <hubert.lacote@gmail.com>
• and contributors