Dear Sofia, I missed to notice this ticket. Sorry for that. A few quick answers now, more details later next week (possibly on the tracker): 1) I believe video decoding, re-encoding, transcoding and streaming are completely separate functionalities compared to video analysis. codoan has a streaming interface to receive (RTP) video streams though. 2) Person detection is something we are examining because of discussions with you at the Madrid meeting. I believe this can be done in the compressed domain to some extend (we already did some experiments on that), but it will not be as accurate as pixel-domain analysis. I do not think that face recognition will be possible in the compressed domain. Originally, codoan was planned to be release only in the FI-WARE test-bed. We are currently discussing of providing also a version that can be installed (or linked into an application), since I believe your use case in SAFECITY is very compelling. Kind regards Peter PS: One more info you have not asked for in your mail but in Madrid: We tested codoan also on Linux and it works fine. > -----Original Message----- > From: Sofia Tsekeridou [mailto:sots at ait.gr] > Sent: Freitag, 20. Juli 2012 16:01 > To: Amon, Peter; Philipp Slusallek [slusallek at dfki.de] > Cc: ab at fi-ppp.eu; fiware-streaming at lists.fi-ware.eu > Subject: RE: [Fiware-streaming] Fiware-streaming Digest, Vol 4, Issue 3 > > Dear Peter, Philipp, all, > > To contribute to the discussion further, I agree with Philipp that if more > detailed information for Fiware GEs is provided and the means to be > notified and easily locate the changes introduced by Fiware developers, > then it will be easier for Use Case projects to assess what further > requirements are posed from their side for a full Fiware-enabled > implementation of their envisioned applications. Of course, this is not > only the case for the Multimedia Analysis GE, as Philipp mentioned, but of > the majority of GEs descriptions and the updating of their status and > plans for future releases. > > Coming now to the specific issues of the Streaming GE WG and their > additional requirements with respect to content analysis and processing > (video, image, audio, 3D), I suggest that I make a first attempt to update > the description in the latest version of the Streaming GE document based > on the updated information that Peter has mentioned on the Multimedia > Analysis GE, both codoan and White Diamond, so as to reflect requirements > that are not currently addressed by the latter by next Tuesday. Then the > Streaming WG as well as Peter or other Fiware representatives could > potentially comment/contribute further. > > On the above, I agree with Philipp that requirements are very much > dependent on the exact algorithms used as well as their > performance/accuracy of detection per different context of use. Thus, such > descriptions and experiments results under which contexts of > use/deployment should be added in the Multimedia Analysis GE. This way we > will better what additional algorithmic requirements we should ask for. I > hereby quote a respective question/request for feedback that I have posted > on the Fusion Forge tool - Peter could you please respond to it? > > ----------------------------- > Date: 2012-07-05 13:37 > Sender: Sofia Tsekeridou > > As discussed during the training week in June 2012 in Madrid, the > following additional functionalities will be desirable to be available in > Fiware Release 1 of the Multimedia Analysis GE: > 1. Video decoding/re-encoding/transcoding/streaming > 2. Person detection (perhaps also face?) > Furthermore, a usage manual for the Multimedia Analysis GE with respect to > algorithm parameters definition, value ranges, and contexts/constraints of > use (e.g. camera settings, lightning conditions, indoor/outdoor settings, > etc.), including performance information (how many camera streams can be > processed in parallel, execution times, computing and memory needs, > accuracy rates, etc.) will be much appreciated along with the 1st Release. > Finally, we have discussed on using the Multimedia Analysis GE on the > SafeCity Gateway during the real-life trials in Madrid in February 2013. > Is it possible to get access to the respective code to do so as well as > installation instructions? If yes, when will it be available? The issue is > that it will be very bandwidth demanding to stream up to 4 HD compressed > camera feeds to the Internet to have them processed by the Multimedia > Analysis GE residing at the Seville testbed. > ------------------------------ > > Many thanks in advance! > > With warm regards, > > Sofia > *************************************************** > Dr. Sofia Tsekeridou > Assistant Professor > Head of Multimedia, Knowledge and Web Technologies Research Area > Athens Information Technology - AIT > 0.8 km Markopoulou Ave. > 19002 Peania, Athens, GREECE > Tel.: +302106682804, Fax: +302106682703 > E-mail: sots at ait.gr > sofia at ieee.org > Web: http://www.ait.gr/ait_web_site/faculty/sots/tsekeridou.html > *************************************************** > > -----Original Message----- > From: ab-bounces at fi-ppp.eu [mailto:ab-bounces at fi-ppp.eu] On Behalf Of > Philipp Slusallek > Sent: Friday, July 20, 2012 3:00 PM > To: Amon, Peter > Cc: ab at fi-ppp.eu; fiware-streaming at lists.fi-ware.eu > Subject: Re: [FI-PPP AB] [Fiware-streaming] Fiware-streaming Digest, Vol 4, > Issue 3 > > Hi, > > Am 20.07.2012 10:53, schrieb Amon, Peter: > > Therefore, let me describe the properties of the "codoan" asset > directly: codoan analyzes H.264/AVC video streams in the compressed domain. > (This is also the reason, why you did not find anyting about the pixel- > domain in the context of codoan). The current features are: > > - Change/event detection (i.e., dectecting in a video stream if > > something happens in the scene) > > - Moving object detection and segmentation (i.e., extraction of shape > > or bounding box for a moving object in a video stream) > > - Foreground/background segmentation (very similar to moving object > > segmentation, since detected objects are regarded as foreground) For the > next release, we plan to implement: > > - Moving object tracking (i.e., matching of segmented objects from > > frame to frame in a video stream) > > > > I thought that the features of this asset were also well explained in > > the two FI-PPP Software Architects weeks (e.g., what can be done in > > the compressed domain and where are the limitations), but maybe I was > > wrong. I will also check, where the desription of our asset on the > > wiki can be improved. I will come back to you on that. (I am afraid > > there is no notification functionality in the wiki.) > > Unfortunately, most people that I talked to seem not to be interested in > this part (except maybe, as an early filter) but they all require analysis > in the pixel domain. So my email was not about this part at all. > > > I believe a major source of confusion is also, that the Multimedia > Analysis GE includes two assets: > > 1) the "codoan" asset (Siemens): an analysis tool for H.264 video > > streams, > > 2) the "White Diamond" asset (Telecom Italia): an Multimedia Analysis > Broker, that orchestrates different analysis components. > > > White Diamond is not planned for Release 1 of FI-WARE, therefore the > > description might not be as concrete as for the other asset. Maybe a > > way to resolve the confusion is to better separate the features of the > > two assets. I will discuss this inside FI-WARE. > > As you say below, its all about avoiding duplication of work and that is > the background of my email. I wanted to know what is coming so we do not > put a lot of money in the OpenCall into something that will be available > in the Release 2. So knowing what will be coming is essential. This seems > to be related to White Diamond but I cannot find anything substantial > about it or to what degree its features would become available as a GE. > > > On my comment for the "Streaming" GE: My only intent was to avoid > > duplication of work. I fully agree with you that analysis tools > > usually address a specific usage area. codoan was origionally designed > > for the surveillance domain (e.g., as addressed by SAFECITY). If you > > believe that the functionality offered by codoan (i.e., as realized as > > low-complexity compressed-domain analysis) is not sufficient for your > > purposes in the content domain (e.g., if you want to do image > > analysis, not video analysis, or if you need analysis of uncompressed > > material), please keep the requirement in the Open Call. However, here > > we should be more precise what we exactly demand. > > What I got from the discussions with SAFECITY in the AB is that their main > requirements are for analysis in the pixel-domain (but they may be able to > take advantage of the compressed domain as a early filter). But maybe I am > wrong. > > They are monitoring a crowded street and as far as I can tell there is > very little you can do in the compressed domain in this situation. As a > result I am still surprised and confused why pixel-domain analysis seems > to have such a low priority in this GE. BTW, the info on the UC > requirements has been available (at various levels of refinement) for > about one years now from the UCs. I just say that to point out that the > communication seems not to work too well here. > > Sorry, that I single you out in this conversation. It is not meant > personally, your mail just trigger a reaction to long standing key issues. > Media (really "image/video"!) analysis is maybe one of the most discussed > topics in the AB (besides streaming) and it seems to highlight a lot of > the things that we need to improve on. > > > On your comment on the "UC backlog": I agree, the search interface is > > not optimal. I also agree that accepted UC tickets should be reflected > > in the wiki. This has been done for codoan, i.e., the "object > > tracking" request from SAFECITY has become an Epic in the wiki already > > some time ago. (Maybe I should not have mentioned this resource in my > > mail.) Non-accepted UC tickets are not reflected. I guess this a topic > > for the AB. I encourage you to discuss this in the AB. > > Good point, thanks! Can we put this on to our agenda for our next meeting? > > > If there are "permission denied" links: I hope such things can be > > resolved easily. I guess this also is something for the AB, if there > > is a general problem. > > Thanks. > > > I am sorry, if I upset you with my mail. I did not mean to. > > No problem. > > > Best, > > Philipp > > > Kind regards > > Peter > > > > > > > > > > > > However, White Diamond is not planned for the first release therefore > > > >> -----Original Message----- > >> From: Philipp Slusallek [mailto:slusallek at dfki.de] > >> Sent: Freitag, 20. Juli 2012 07:43 > >> To: Amon, Peter > >> Cc: Sofia Tsekeridou; fiware-streaming at lists.fi-ware.eu; ab at fi-ppp.eu > >> Subject: Re: [Fiware-streaming] Fiware-streaming Digest, Vol 4, Issue > >> 3 > >> > >> Hi, > >> > >> Thank you very much, Peter. > >> > >> This brief discussion about finding out more what is in a GE is > >> touching on fundamental issues of collaboration between FI-WARE, the > >> AB, and the UC projects -- for a topic that is of high importance to > >> many UCs and which has been discussed extensively in the AB. So allow > >> me to raise this conversation to the attention of the AB. > >> > >> > >> In your answer to my question trying to find out what is coming in > >> terms of pixel-level media analysis you simply refer me to the Wiki > >> in general terms. I happen to know about the information sources in > >> the FI-PPP. As we all know they are not perfect. The Product Vision > >> is extremely high level and promises the moon, but as we found out in > >> Madrid, we seem to be getting much less. Trying to find out what this > >> is, is important as other work in the UCs will be based on what we > >> will get and not on what is in the vision. > >> > >> Let me tell you that I did read all the documentation on multimedia > >> analysis in quite some detail a while ago (it's an area that is > >> important to our work). After our discussion in Madrid, my question > >> has been more than fair. There you essentially said that pixel-domain > >> processing would not come and we only found out later that there will > >> be something but that there is no information on this yet on the Wiki. > >> > >> Since there is no way I can get notified (as far as I know) when Wiki > >> specifications change for GEs that are relevant to me, I expect a > >> reasonable answer to my question and not a "you find it on the Web". > >> How about simply sending us a URL to the relevant sections, instead > >> of listing all the high-level info we all know perfectly well. As you > >> see, I am getting a bit upset here -- and I apologize for this. > >> > >> Back to my questions: You have criticized the definition of our new > >> GEs in the sense that they cover things that will be coming in the > >> existing GEs. This would be fair -- IF there is enough info about > >> what is coming so that we can even know about it and plan > >> accordingly. We have very concrete use cases that we must implement > >> and so need very concrete info about what will be offered. Simply > >> saying "there will be moving object segmentation" is like saying > >> nothing in the context of multimedia analysis. As we all know, there > >> are tons of algorithms and each one only works for some specific > >> scenarios. Without knowing in more detail what type of algorithms is > >> being used and how this can be tweaked to work with various types of > >> objects and environments, it is mostly useless to someone who wants to > apply your GE. > >> > >> In this concrete case: In response to your email, this morning I did > >> search through all of the parts of the FI-WARE Wiki relevant to > >> multimedia analysis in search for more details about pixel-level > >> analysis. I found NOTHING! (that I had not seen before and which was > >> not adequate). OK, maybe I missed it -- which is even not unlikely. > >> But that is exactly why sending me a concrete link as an answer would > >> have been so much more helpful. > >> > >> So may I politely ask again for more detailed information about what > >> will be coming in terms of pixel-level analysis? > >> > >> As a side note: It often has helped to look at the documentation of > >> the base assets to guess what will be supported. In this context, I > "love" > >> the description of the "baseline assets" for multimedia analysis: > >> codoan is described as a proprietary tool and it refers to the vision > >> for more details (which are obviously not there) and White Diamond > >> has two links to more details for both of which I still get a > "permission denied" > >> error. You are certainly not making it easy to find out details. > >> > >> Also, you mention that there is info in the tickets about changes to > >> the GE. May I politely ask that accepted requests that modify what is > >> in the GE specs get integrated into the description of the GE with > >> links back to their discussion in the tickets (we are on the Web > >> after all). I would even strongly suggest to link to refused requests > >> such that people do not ask for these things again. You cannot > >> reasonably expect people to search the ticket data base (with its > >> sub-optimal search interface) and wade through tickets to find out > >> about things that may or will be in a GE. > >> > >> > >> Best, > >> > >> Philipp > >> > >> > >> Am 19.07.2012 22:19, schrieb Amon, Peter: > >>> a good source for such a description is the Epic and Feature > >>> overview > >> for this GE, to be found under "Materializing the FI-WARE vision" on > >> the FI-WARE Wiki. Since recently also "Release 0" Features were added > >> (like for all FI-WARE GEs), this should give a complete picture of > >> the functionality for the GE. > >>> > >>> An additional source is the UC backlog, especially those tickets > >>> labeled > >> "Accepted for Inclusion in FI-WARE Backlog", since these indicate > >> future planning for this GE. Here, requirements from UC projects > >> (e.g., provided by Sofia in the case of MMA) were discussed and > >> included into the development process for FI-WARE GEs. > >>> > >>> Finally, "FI-WARE Product Vision" and "FI-WARE Architecture and Open > >> Specifications" (both on the FI-WARE Wiki) give a more general > >> overview for FI-WARE GEs. However, in the case of the Multimedia > >> Analysis GE it includes properties of both, the "codoan" asset by > >> Siemens and the "White Diamond" asset from Telecom Italia. > >>> > >>> Kind regards > >>> Peter > >>> > >>> > >>>> -----Original Message----- > >>>> From: fiware-streaming-bounces at lists.fi-ware.eu [mailto:fiware- > >> streaming- > >>>> bounces at lists.fi-ware.eu] On Behalf Of Philipp Slusallek > >>>> Sent: Donnerstag, 19. Juli 2012 13:04 > >>>> To: Sofia Tsekeridou > >>>> Cc: fiware-streaming at lists.fi-ware.eu > >>>> Subject: Re: [Fiware-streaming] Fiware-streaming Digest, Vol 4, > >>>> Issue 3 > >>>> > >>>> Hi, > >>>> > >>>> Am 19.07.2012 12:51, schrieb Sofia Tsekeridou: > >>>>> With respect to the moving object segmentation and tracking > >>>>> capabilities of the Multimedia Analysis GE, is it possible to also > >>>>> extract information about the movement? E.g. direction of > >>>>> movement, speed, or generally movement statistics along with e.g. > >>>>> shape of the moving object, etc.? Is it also possible to identify > >>>>> rigid versus non-rigid moving objects (e.g. cars versus persons)? > >>>>> These may be better evaluated in the pixel domain than in the > compressed domain. > >>>>> What do you think? > >>>> > >>>> Asked differently, is there a more detailed description of what to > >>>> expect in this GE (as far as I remember there was no description of > >> what > >>>> is to come). This is relevant in the context that two of the new > >>>> GEs will need high-performance, real-time, highly flexible, > >>>> pixel-based analysis even on mobile platforms (at least for Augmented > Reality). > >>>> > >>>> > >>>> Best, > >>>> > >>>> Philipp > >>>> > >>>>> Many thanks, > >>>>> > >>>>> Sofia > >>>>> > >>>>> *************************************************** > >>>>> Dr. Sofia Tsekeridou > >>>>> Assistant Professor > >>>>> Head of Multimedia, Knowledge and Web Technologies Research Area > >>>>> Athens Information Technology - AIT > >>>>> 0.8 km Markopoulou Ave. > >>>>> 19002 Peania, Athens, GREECE > >>>>> Tel.: +302106682804, Fax: +302106682703 > >>>>> E-mail: sots at ait.gr > >>>>> sofia at ieee.org > >>>>> Web: http://www.ait.gr/ait_web_site/faculty/sots/tsekeridou.html > >>>>> *************************************************** > >>>>> > >>>>> ________________________________________ > >>>>> From: fiware-streaming-bounces at lists.fi-ware.eu [fiware-streaming- > >>>> bounces at lists.fi-ware.eu] on behalf of fiware-streaming- > >> request at lists.fi- > >>>> ware.eu [fiware-streaming-request at lists.fi-ware.eu] > >>>>> Sent: 19 July 2012 13:00 > >>>>> To: fiware-streaming at lists.fi-ware.eu > >>>>> Subject: Fiware-streaming Digest, Vol 4, Issue 3 > >>>>> > >>>>> Send Fiware-streaming mailing list submissions to > >>>>> fiware-streaming at lists.fi-ware.eu > >>>>> > >>>>> To subscribe or unsubscribe via the World Wide Web, visit > >>>>> http://lists.fi-ware.eu/listinfo/fiware-streaming > >>>>> or, via email, send a message with subject or body 'help' to > >>>>> fiware-streaming-request at lists.fi-ware.eu > >>>>> > >>>>> You can reach the person managing the list at > >>>>> fiware-streaming-owner at lists.fi-ware.eu > >>>>> > >>>>> When replying, please edit your Subject line so it is more > >>>>> specific than "Re: Contents of Fiware-streaming digest..." > >>>>> > >>>>> > >>>>> Today's Topics: > >>>>> > >>>>> 1. Re: FI-WARE streaming updates (Amon, Peter) > >>>>> > >>>>> > >>>>> ------------------------------------------------------------------ > >>>>> ---- > >>>>> > >>>>> Message: 1 > >>>>> Date: Wed, 18 Jul 2012 23:15:44 +0200 > >>>>> From: "Amon, Peter" <p.amon at siemens.com> > >>>>> To: Mischler Denis <denis.mischler at technicolor.com>, "ab at fi-ppp.eu" > >>>>> <ab at fi-ppp.eu>, "fiware-streaming at lists.fi-ware.eu" > >>>>> <fiware-streaming at lists.fi-ware.eu> > >>>>> Subject: Re: [Fiware-streaming] FI-WARE streaming updates > >>>>> Message-ID: > >>>>> > >>>> > >> <EDDE8D0FEF8B6B4AA50AC1AC8BB744DE03A409DE3A at DEMCHP99E35MSX.ww902.siem > >> ens.n > >>>> et> > >>>>> > >>>>> Content-Type: text/plain; charset="us-ascii" > >>>>> > >>>>> Dear Denis, all, > >>>>> > >>>>> thanks for the new version of the document! > >>>>> > >>>>> Just one comment on FIWARE.EPIC.STEAMING.GENERIC_PROCESSING: > >>>>> "Moving > >>>> object segmentation" (and to some extent also > >>>> "Foreground/background estimation" and "Moving object > >>>> identification") is supported by the > >> codoan > >>>> asset realizing part of the Multimedia Analysis GE. "Moving object > >>>> tracking" is planned for future releases of this GE. Therefore, I > >>>> doubt that we should ask for these functionalities in an Open Call. > >>>>> > >>>>> My proposal: > >>>>> - Remove "Moving object segmentation" and "Foreground/background > >>>> estimation" > >>>>> - Remove "Moving object identification (person, car, animal, > >>>>> suitcase, > >>>> etc.) and tracking" and add instead "Object identification (person, > >>>> car, animal, suitcase, etc.)" > >>>>> > >>>>> Thanks for considering the proposal. > >>>>> > >>>>> Kind regards > >>>>> Peter > >>>>> > >>>>> > >>>>> ________________________________ > >>>>> From: fiware-streaming-bounces at lists.fi-ware.eu [mailto:fiware- > >>>> streaming-bounces at lists.fi-ware.eu] On Behalf Of Mischler Denis > >>>>> Sent: Mittwoch, 18. Juli 2012 11:31 > >>>>> To: ab at fi-ppp.eu; fiware-streaming at lists.fi-ware.eu > >>>>> Subject: [Fiware-streaming] FI-WARE streaming updates > >>>>> > >>>>> Dear all, > >>>>> here a few updates in the streaming document (track changes > >>>>> mode) > >>>>> > >>>>> - Added requirements about multi view (FI-CONTENT Telecom > Italia) > >>>>> > >>>>> - Charging and Payment : used the word "Service Data logging". > >>>> (only a vocabulary correction) > >>>>> > >>>>> - Added a storage EPIC to be decided about (here or in the > >> Storage > >>>> Enabler) and completed if "yes". > >>>>> Regards > >>>>> Denis Mischler > >>>>> -------------- next part -------------- An HTML attachment was > >>>>> scrubbed... > >>>>> URL: <http://lists.fi-ware.eu/private/fiware- > >>>> streaming/attachments/20120718/dfdb0349/attachment-0001.html> > >>>>> > >>>>> ------------------------------ > >>>>> > >>>>> _______________________________________________ > >>>>> Fiware-streaming mailing list > >>>>> Fiware-streaming at lists.fi-ware.eu > >>>>> http://lists.fi-ware.eu/listinfo/fiware-streaming > >>>>> > >>>>> > >>>>> End of Fiware-streaming Digest, Vol 4, Issue 3 > >>>>> ********************************************** > >>>>> > >>>>> > >>>>> _______________________________________________ > >>>>> Fiware-streaming mailing list > >>>>> Fiware-streaming at lists.fi-ware.eu > >>>>> http://lists.fi-ware.eu/listinfo/fiware-streaming > >>>>> > >>>> > >>>> > >>>> -- > >>>> > >>>> ------------------------------------------------------------------- > >>>> ---- > >> -- > >>>> Deutsches Forschungszentrum für Künstliche Intelligenz (DFKI) GmbH > >>>> Trippstadter Strasse 122, D-67663 Kaiserslautern > >>>> > >>>> Gesch?ftsführung: > >>>> Prof. Dr. Dr. h.c. mult. Wolfgang Wahlster (Vorsitzender) > >>>> Dr. Walter Olthoff > >>>> Vorsitzender des Aufsichtsrats: > >>>> Prof. Dr. h.c. Hans A. Aukes > >>>> > >>>> Sitz der Gesellschaft: Kaiserslautern (HRB 2313) > >>>> USt-Id.Nr.: DE 148646973, Steuernummer: 19/673/0060/3 > >>>> ------------------------------------------------------------------- > >>>> ---- > >> --- > >>>> - > >> > >> > >> -- > >> > >> --------------------------------------------------------------------- > >> ---- Deutsches Forschungszentrum für Künstliche Intelligenz (DFKI) > >> GmbH Trippstadter Strasse 122, D-67663 Kaiserslautern > >> > >> Gesch?ftsführung: > >> Prof. Dr. Dr. h.c. mult. Wolfgang Wahlster (Vorsitzender) > >> Dr. Walter Olthoff > >> Vorsitzender des Aufsichtsrats: > >> Prof. Dr. h.c. Hans A. Aukes > >> > >> Sitz der Gesellschaft: Kaiserslautern (HRB 2313) > >> USt-Id.Nr.: DE 148646973, Steuernummer: 19/673/0060/3 > >> --------------------------------------------------------------------- > >> ----- > >> - > > > > _______________________________________________ > > Ab mailing list > > Ab at fi-ppp.eu > > http://lists.fi-ppp.eu/mailman/listinfo/ab > > >
You can get more information about our cookies and privacy policies clicking on the following links: Privacy policy Cookies policy