Hi Sakari,
On 12/10/2011 11:44 AM, Sakari Ailus wrote:
> On Wed, Dec 07, 2011 at 12:09:09PM +0100, Sylwester Nawrocki wrote:
>> On 12/06/2011 05:27 PM, Sylwester Nawrocki wrote:
>>>>> +
>>>>> <entry><constant>V4L2_METERING_MODE_CENTER_WEIGHTED</constant> </entr
>>>>> y> + <entry>Average the light information coming from the entire scene
>>>>> +giving priority to the center of the metered area.</entry>
>>>>> + </row>
>>>>> + <row>
>>>>> + <entry><constant>V4L2_METERING_MODE_SPOT</constant> </entry>
>>>>> + <entry>Measure only very small area at the cent-re of the
>>>>> scene.</entry> + </row>
>>>>> + </tbody>
>>>>
>>>> For the last two cases, would it also make sense to specify the center of the
>>>> weighted area and the spot location ?
>>>
>>> Yes, that's quite basic requirement as well.. A means to determine the
>>> location would be also needed for some auto focus algorithms.
>>>
>>> Additionally for V4L2_METERING_MODE_CENTER_WEIGHTED it's also needed to
>>> specify the size of the area (width/height).
>>>
>>> What do you think about defining new control for passing pixel position,
>>> i.e. modifying struct v4l2_ext_control to something like:
>>>
>>> struct v4l2_ext_control {
>>> __u32 id;
>>> __u32 size;
>>> __u32 reserved2[1];
>>> union {
>>> __s32 value;
>>> __s64 value64;
>>> struct v4l2_point position;
>>> char *string;
>>> };
>>> } __attribute__ ((packed));
>>>
>>> where:
>>>
>>> struct v4l2_point {
>>> __s32 x;
>>> __s32 y;
>>> };
>>
>> Hmm, that won't work since there is no way to handle the min/max/step for
>> more than one value. Probably the selection API could be used for specifying
>> the metering rectangle, or just separate controls for x, y, width, height.
>> Since we need to specify only locations for some controls and a rectangle for
>> others, probably separate controls would be more suitable.
>
> I prefer the use of the selection API over controls. Also consider you may
> have multiple areas for metering. Also the areas may well be different for
> focus, white balance and exxposure --- they often actually are.
Yes, AFAIR Laurent expressed similar opinion on that [1].
I talked to Tomasz and Marek about configuring of the metering areas and we came
to conclusion that we could designate a group of targets for exposure, auto-focus,
etc. Define a base and target pool size so it is possible to define multiple
rectangles. And single point (pixel) would be described by struct v4l2_rect with
width=1, height=1.
Then we would probably need to be able to enumerate the targets somehow.
[1] http://www.spinics.net/lists/linux-media/msg41215.html
--
Thanks,
Sylwester
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
