path: root/docs/manual.md

% libuca -- A Unified Camera Access Interface
% Matthias Vogelgesang [matthias.vogelgesang@kit.edu]

libuca is a light-weight camera abstraction library, focused on scientific
cameras used at the ANKA synchrotron.

# Quickstart

## Installation

Before installing `libuca` itself, you should install any drivers and SDKs
needed to access the cameras you want to access through `libuca`. Now you have
two options: install pre-built packages or build from source.

### Installing packages

Packages for the core library and all plugins are currently provided for
openSUSE. To install them run `zypper`:

    sudo zypper in libuca-x.y.z-x86_64.rpm
    sudo zypper in uca-plugin-*.rpm

To install development files such as headers, you have to install the
`libuca-x.y.z-devel.rpm` package.

### Building from source

Building the library and installing from source is simple and straightforward.
Make sure you have

* CMake,
* a C compiler,
* GLib and GObject development libraries and
* necessary camera SDKs

installed.

For the base system, install

    [Debian] sudo apt-get install libglib2.0 cmake gcc
    [openSUSE] sudo zypper in glib2-devel cmake gcc

In case you want to use the graphical user interface you also need the Gtk+
development libraries:

    [Debian] sudo apt-get install libgtk+2.0-dev
    [openSUSE] sudo zypper in gtk2-devel

To generate bindings for third-party languages, you have to install

    [Debian] sudo apt-get install gobject-introspection
    [openSUSE] sudo zypper in gobject-introspection-devel


#### Fetching the sources

Untar the distribution

    untar xfz libuca-x.y.z.tar.gz

or clone the repository

    git clone http://ufo.kit.edu/git/libuca

and create a new, empty build directory inside:

    cd libuca/
    mkdir build


#### Configuring and building

Now you need to create the Makefile with CMake. Go into the build directory and
point CMake to the `libuca` top-level directory:

    cd build/
    cmake ..

As long as the last line reads "Build files have been written to", the
configuration stage is successful. In this case you can build `libuca` with

    make

and install with

    sudo make install

If an _essential_ dependency could not be found, the configuration stage will stop
and build files will not be written. If a _non-essential_ dependency (such as a
certain camera SDK) is not found, the configuration stage will continue but that
particular camera support not built.

If you want to customize the build process you can pass several variables to
CMake:

    cmake .. -DPREFIX=/usr -DLIBDIR=/usr/lib64

The former tells CMake to install into `/usr` instead of `/usr/local` and the
latter that we want to install the libraries and plugins into the `lib64` subdir
instead of the default `lib` subdir as it is common on SUSE systems.

#### Building this manual

Make sure you have [Pandoc][] installed. With Debian/Ubuntu this can be achieved
with

    sudo apt-get install pandoc

Once done, go into `docs/` and type

    make [all|pdf|html]

[Pandoc]: http://johnmacfarlane.net/pandoc/


## First look at the API

The API for accessing cameras is straightforward. First you need to include the
necessary header files:

~~~ {.c}
#include <glib-object.h>
#include <uca/uca-plugin-manager.h>
#include <uca/uca-camera.h>
~~~

Then you need to setup the type system:

~~~ {.c}
int
main (int argc, char *argv[])
{
    UcaPluginManager *manager;
    UcaCamera *camera;
    GError *error = NULL; /* this _must_ be set to NULL */

    g_type_init ();
~~~

Now you can instantiate new camera _objects_. Each camera is identified by a
human-readable string, in this case we want to access any pco camera that is
supported by [libpco][]. To instantiate a camera we have to create a plugin
manager first:

~~~ {.c}
    manager = uca_plugin_manager_new ();
    camera = uca_plugin_manager_get_camera (manager, "pco", &error);
~~~

Errors are indicated with a returned value `NULL` and `error` set to a value
other than `NULL`:

~~~ {.c}
    if (camera == NULL) {
        g_error ("Initialization: %s", error->message);
        return 1;
    }
~~~

You should always remove the [reference][gobject-references] from the camera
object when not using it in order to free all associated resources:

~~~ {.c}
    g_object_unref (camera);
    return 0;
}
~~~

Compile this program with

    cc `pkg-config --cflags --libs libuca glib-2.0` foo.c -o foo

Now, run `foo` and verify that no errors occur.


[libpco]: http://ufo.kit.edu/repos/libpco.git/
[gobject-references]: http://developer.gnome.org/gobject/stable/gobject-memory.html#gobject-memory-refcount


### Grabbing frames

To synchronously grab frames, first start the camera:

~~~ {.c}
    uca_camera_start_recording (camera, &error);
    g_assert_no_error (error);
~~~

Now, you have to allocate a suitably sized buffer and pass it to
`uca_camera_grab`.

~~~ {.c}
    gpointer buffer = g_malloc0 (640 * 480 * 2);

    uca_camera_grab (camera, buffer, &error);
~~~

You have to make sure that the buffer is large enough by querying the size of
the region of interest and the number of bits that are transferred.


### Getting and setting camera parameters

Because camera parameters vary tremendously between different vendors and
products, they are realized with so-called GObject _properties_, a mechanism
that maps string keys to typed and access restricted values. To get a value, you
use the `g_object_get` function and provide memory where the result is stored:

~~~ {.c}
    guint roi_width;
    gdouble exposure_time;

    g_object_get (G_OBJECT(camera),
                  "roi-width", &roi_width,
                  "exposure-time", &exposure_time,
                  /* The NULL marks the end! */
                  NULL
                  );

    g_print ("Width of the region of interest: %d\n", roi_width);
    g_print ("Exposure time: %3.5s\n", exposure_time);
~~~

In a similar way, properties are set with `g_object_set`:

~~~ {.c}
    guint roi_width = 512;
    gdouble exposure_time = 0.001;

    g_object_set (G_OBJECT (camera),
                  "roi-width", roi_width,
                  "exposure-time", exposure_time,
                  NULL);
~~~

Each property can be associated with a physical unit. To query for the unit call
`uca_camera_get_unit` and pass a property name. The function will then return a
value from the `UcaUnit` enum.

Several essential camera parameters _must_ be implemented by all cameras. To get
a list of them consult the API reference for [`UcaCamera`][ucacam-ref]. For
camera specific parameters you need to consult the corresponding API reference
for `UfoFooCamera`. The latest nightly built reference can be found
[here][libuca-reference].

[ucacam-ref]: http://ufo.kit.edu/extra/libuca/reference/UcaCamera.html#UcaCamera.properties
[libuca-reference]: http://ufo.kit.edu/extra/libuca/reference/


# Supported cameras

The following cameras are supported:

* pco.edge, pco.dimax, pco.4000 (all CameraLink) via [libpco][]. You need to
  have the SiliconSoftware frame grabber SDK with the `menable` kernel module
  installed.
* PhotonFocus
* Pylon
* UFO Camera developed at KIT/IPE.

## Property documentation

* [mock][mock-doc]
* [pco][pco-doc]
* [PhotonFocus][pf-doc]
* [Ufo Camera][ufo-doc]

[mock-doc]: mock.html
[pco-doc]: pco.html
[pf-doc]: pf.html
[ufo-doc]: ufo.html


# More API

In the [last section][], we had a quick glance over the basic API used to
communicate with the camera. Now we will go into more detail.

## Instantiating cameras

We have already seen how to instantiate a camera object from a name. If you have
more than one camera connected to a machine, you will most likely want the user
decide which to use. To do so, you can enumerate all camera strings with
`uca_plugin_manager_get_available_cameras`:

~~~ {.c}
    GList *types;

    types = uca_camera_get_available_cameras (manager);

    for (GList *it = g_list_first; it != NULL; it = g_list_next (it))
        g_print ("%s\n", (gchar *) it->data);

    /* free the strings and the list */
    g_list_foreach (types, (GFunc) g_free, NULL);
    g_list_free (types);
~~~

[last section]: #first-look-at-the-api


## Errors

All public API functions take a location of a pointer to a `GError` structure as
a last argument. You can pass in a `NULL` value, in which case you cannot be
notified about exceptional behavior. On the other hand, if you pass in a
pointer to a `GError`, it must be initialized with `NULL` so that you do not
accidentally overwrite and miss an error occurred earlier.

Read more about `GError`s in the official GLib
[documentation][GError].

[GError]: http://developer.gnome.org/glib/stable/glib-Error-Reporting.html


## Recording

Recording frames is independent of actually grabbing them and is started with
`uca_camera_start_recording`. You should always stop the recording with
`ufo_camera_stop_recording` when you finished. When the recording has started,
you can grab frames synchronously as described earlier. In this mode, a block to
`uca_camera_grab` blocks until a frame is read from the camera. Grabbing might
block indefinitely, when the camera is not functioning correctly or it is not
triggered automatically.


## Triggering

`libuca` supports three trigger modes through the "trigger-mode" property:

1. `UCA_CAMERA_TRIGGER_AUTO`: Exposure is triggered by the camera itself.
2. `UCA_CAMERA_TRIGGER_INTERNAL`: Exposure is triggered via software.
3. `UCA_CAMERA_TRIGGER_EXTERNAL`: Exposure is triggered by an external hardware
   mechanism.

With `UCA_CAMERA_TRIGGER_INTERNAL` you have to trigger with
`uca_camera_trigger`:

~~~ {.c}
    /* thread A */
    g_object_set (G_OBJECT (camera),
                  "trigger-mode", UCA_CAMERA_TRIGGER_INTERNAL,
                  NULL);

    uca_camera_start_recording (camera, NULL);
    uca_camera_grab (camera, &buffer, NULL);
    uca_camera_stop_recording (camera, NULL);

    /* thread B */
    uca_camera_trigger (camera, NULL);
~~~


## Grabbing frames asynchronously

In some applications, it might make sense to setup asynchronous frame
acquisition, for which you will not be blocked by a call to `libuca`:

~~~ {.c}
static void
callback (gpointer buffer, gpointer user_data)
{
    /*
     * Do something useful with the buffer and the string we have got.
     */
}

static void
setup_async (UcaCamera *camera)
{
    gchar *s = g_strdup ("lorem ipsum");

    g_object_set (G_OBJECT (camera),
                  "transfer-asynchronously", TRUE,
                  NULL);

    uca_camera_set_grab_func (camera, callback, s);
    uca_camera_start_recording (camera, NULL);

    /*
     * We will return here and `callback` will be called for each newo
     * new frame.
     */
}
~~~


# Bindings

Since version 1.1, libuca generates GObject introspection meta data if
`g-ir-scanner` and `g-ir-compiler` can be found. When the XML description
`Uca-x.y.gir` and the typelib `Uca-x.y.typelib` are installed, GI-aware
languages can access libuca and create and modify cameras, for example in
Python:

~~~ {.python}
from gi.repository import Uca

pm = Uca.PluginManager()

# List all cameras
print(pm.get_available_cameras())

# Load a camera
cam = pm.get_camerav('pco', [])

# You can read and write properties in two ways
cam.set_properties(exposure_time=0.05)
cam.props.roi_width = 1024
~~~

Note, that the naming of classes and properties depends on the GI implementation
of the target language. For example with Python, the namespace prefix `uca_`
becomes the module name `Uca` and dashes separating property names become
underscores.

Integration with Numpy is relatively straightforward. The most important thing
is to get the data pointer from a Numpy array to pass it to `uca_camera_grab`:

~~~ {.python}
import numpy as np

def create_array_from(camera):
    """Create a suitably sized Numpy array and return it together with the
    arrays data pointer"""
    bits = camera.props.sensor_bitdepth
    dtype = np.uint16 if bits > 8 else np.uint8
    a = np.zeros((cam.props.roi_height, cam.props.roi_width), dtype=dtype)
    return a, a.__array_interface__['data'][0]

# Suppose 'camera' is a already available, you would get the camera data like
# this:
a, buf = create_array_from(camera)
camera.start_recording()
camera.grab(buf)

# Now data is in 'a' and we can use Numpy functions on it
print(np.mean(a))

camera.stop_recording()
~~~


# Integrating new cameras

A new camera is integrated by [sub-classing][] `UcaCamera` and implement all
virtual methods. The simplest way is to take the `mock` camera and
rename all occurences. Note, that if you class is going to be called `FooBar`,
the upper case variant is `FOO_BAR` and the lower case variant is `foo_bar`.

In order to fully implement a camera, you need to override at least the
following virtual methods:

* `start_recording`: Take suitable actions so that a subsequent call to
  `grab` delivers an image or blocks until one is exposed.
* `stop_recording`: Stop recording so that subsequent calls to `grab`
  fail.
* `grab`: Return an image from the camera or block until one is ready.

## Asynchronous operation

When the camera supports asynchronous acquisition and announces it with a true
boolean value for `"transfer-asynchronously"`, a mechanism must be setup up
during `start_recording` so that for each new frame the grab func callback is
called.

## Cameras with internal memory

Cameras such as the pco.dimax record into their own on-board memory rather than
streaming directly to the host PC. In this case, both `start_recording` and
`stop_recording` initiate and end acquisition to the on-board memory. To
initiate a data transfer, the host calls `start_readout` which must be suitably
implemented. The actual data transfer happens either with `grab` or
asynchronously.


[sub-classing]: http://developer.gnome.org/gobject/stable/howto-gobject.html


# Tools

Several tools are available to ensure `libuca` works as expected. All of them
are located in `build/test/` and some of them are installed with `make
installed`.

## `uca-grab` -- grabbing frames

Grab with frames with

    $ uca-grab --num-frames=10 camera-model

store them on disk as `frames.tif` if `libtiff` is installed, otherwise as
`frame-00000.raw`, `frame-000001.raw`. The raw format is a memory dump of the
frames, so you might want to use [ImageJ][] to view them. You can also specify
the output filename or filename prefix with the ``-o/--output`` option:

    $ uca-grab -n 10 --output=foobar.tif camera-model

Instead of reading exactly _n_ frames, you can also specify a duration in
fractions of seconds:

    $ uca-grab --duration=0.25 camera-model

[ImageJ]: http://rsbweb.nih.gov/ij/


## `uca-camera-control` -- simple graphical user interface

Shows the frames and displays them on screen. Moreover, you can change the
camera properties in a side pane.

## `uca-benchmark` -- check bandwidth

Measure the memory bandwidth by taking subsequent frames and averaging the
grabbing time:

    $ ./benchmark mock
    # --- General information ---
    # Sensor size: 640x480
    # ROI size: 640x480
    # Exposure time: 0.000010s
    # type      n_frames  n_runs    frames/s        MiB/s
      sync      100       3         29848.98        8744.82
      async     100       3         15739.43        4611.16


# The GObject Tango device

UcaDevice is a generic Tango Device that wraps `libuca` in order to make libuca controlled cameras available as Tango devices.

## Architecture

UcaDevice is derived from GObjectDevice and adds libuca specific features like start/stop recording etc. 

The most important feature is _acquisition control_. UcaDevice is responsible for controlling acquisition of images from libuca. The last aquired image can be accessed by reading attribute `SingleImage`.

UcaDevice is most useful together with ImageDevice. If used together, UcaDevice sends each aquired image to ImageDevice, which in turn does configured post-processing like flipping, rotating or writing the image to disk.

## Attributes

Besides the dynamic attributes translated from libuca properties UcaDevice has the following attributes:

* `NumberOfImages (Tango::DevLong)`: how many images should be acquired? A value of -1 means unlimited _(read/write)_
* `ImageCounter (Tango::DevULong)`: current number of acquired images _(read-only)_
* `CameraName (Tango::DevString)`: name of libuca object type _(read-only)_
* `SingleImage (Tango::DevUChar)`: holds the last acquired image

## Acquisition Control

In UcaDevice acquisition control is mostly implemented by two `yat4tango::DeviceTasks`: _AcquisitionTask_ and _GrabTask_. _GrabTask_'s only responsibility is to call `grab` on `libuca` synchronously and post the data on to AcquisitionTask. 

_AcquisitionTask_ is responsible for starting and stopping GrabTask and for passing image data on to `ImageDevice` (if exisiting) and to `UcaDevice` for storage in attribute `SingleImage`. It counts how many images have been acquired and checks this number against the configured `NumberOfImages`. If the desired number is reached, it stops GrabTask, calls `stop_recording` on `libuca` and sets the Tango state to STANDBY.

## Plugins

As different cameras have different needs, plugins are used for special implementations. Plugins also makes UcaDevice and Tango Servers based on it more flexible and independent of libuca implementation. 

## Pco

The Pco plugin implements additional checks when writing ROI values.

## Pylon

The Pylon plugin sets default values for `roi-width` and `roi-height` from libuca properties `roi-width-default` and `roi-height-default`.

## Installation 

Detailed installation depends on the manifestation of UcaDevice. <br />
All manifestations depend on the following libraries:

* YAT
* YAT4Tango
* Tango
* GObjectDevice
* ImageDevice

## Build

    export PKG_CONFIG_PATH=/usr/lib/pkgconfig
    export PYLON_ROOT=/usr/pylon
    export LD_LIBRARY_PATH=$PYLON_ROOT/lib64:$PYLON_ROOT/genicam/bin/Linux64_x64
    git clone git@iss-repo:UcaDevice.git
    cd UcaDevice
    mkdir build
    cd build
    cmake ..
    make


## Setup in Tango Database / Jive

Before `ds_UcaDevice` can be started, it has to be registered manually in the Tango database. With `Jive` the following steps are necessary:

[1] Register Server <br />
Menu _Tools_ &#8594; Server Wizard <br />
Server name &#8594; ds_UcaDevice <br />
Instance name &#8594; my_server  _(name can be chosen freely)_  <br />
Next <br />
Cancel

[2] Register Classes and Instances <br />
In tab _Server_: context menu on ds_UcaDevice &#8594; my_server &#8594; Add Class <br />
Class: UcaDevice <br />
Devices: `iss/name1/name2` <br />
Register server <br />
same for class ImageDevice

[3] Start server

    export TANGO_HOST=anka-tango:100xx
    export UCA_DEVICE_PLUGINS_DIR=/usr/lib(64)
    ds_UcaDevice pco my_server

[4] Setup properties for UcaDevice <br />
context menu on device &#8594; Device wizard <br />
Property StorageDevice: _address of previously registered ImageDevice instance_

[5] Setup properties for ImageDevice <br />
context menu on device &#8594; Device wizard <br />
PixelSize: how many bytes per pixel for the images of this camera? <br />
GrabbingDevice: _address of previously registered UcaDevice instance_

[6] Finish <br />
restart ds_UcaDevice

## FAQ

_UcaDevice refuses to start up...?_ <br />
Most likely there is no instance registered for class UcaDevice. Register an instance for this class and it should work.

_Why does UcaDevice depend on ImageDevice?_ <br />
UcaDevice pushes each new Frame to ImageDevice. Polling is not only less efficient but also prone to errors, e.g. missed/double frames and so on. Perhaps we could use the Tango-Event-System here! 

## Open Questions, Missing Features etc.

_Why do we need to specify `Storage` for UcaDevice and `GrabbingDevice` for ImageDevice?_ <br />
ImageDevice needs the Tango-Address of UcaDevice to mirror all Attributes and Commands and to forward them to it. UcaDevice needs the Tango-Address of ImageDevice to push a new frame on reception. A more convenient solution could be using conventions for the device names, e.g. of the form `$prefix/$instance_name/uca` and `$prefix/$instance_name/image`. That way we could get rid of the two Device-Properties and an easier installation without the process of registering the classes and instances in `Jive`.

_Why does UcaDevice dynamically link to GObjectDevice?_ <br />
There is no good reason for it. Packaging and installing would be easier if we linked statically to `GObjectDevice` because we would hide this dependency. Having a separate `GObjectDevice` is generally a nice feature to make `GObjects` available in Tango. However, there is currently no GObjectDevice in use other than in the context of UcaDevice.

_Why must the plugin name be given as a command line parameter instead of a Device-Property?_ <br />
There is no good reason for it. UcaDevice would be easier to use, if the plugin was configured in the Tango database as a Device-Property for the respective server instance.