Merken

Image recognition and camera positioning with OpenCV. A tourist guide application.

Zitierlink des Filmsegments
Embed Code

Automatisierte Medienanalyse

Beta
Erkannte Entitäten
Sprachtranskript
few good morning my name is that this is a today I'm going to talk about image recognition and camera positioning with be open to the arteries guide mediation I work for Bjorkman solutions room we developed software solutions for managing and publishing just assume that they're using the open-source software based on Linux and he major
animal cell is a few days in evolution the companies invest a lot of resources in these in research field story well for OK the the main issues in image information on the following 1st of all that we have we have to implement a human ability this is a very challenging that the majors to compare can be distracted and oriented in different ways so we have to detect several features and the algorithm of detection would be skin by for a scope we caffeine too user related to user a major commission to look like all images so let's see why we cannot use it to this is a typical images taken with a
smartphone horrible glass and we have to recognize the picture in the center and the and all my boss the vector and the In William major so we can not to define library of from images so he tries to to match all the features in our image and not only the features features so you recognize the Roma we our of featuring the center or not or other things so we have to find the other strategies let's try to
understand what could be the the
best candidate as a features let's take a rectangle
for example we can find 3 possible areas that's only is that flat surfaces if we move from the IRA in any direction Dick context of this query would not change in aspect aspects so we can we can not the locality so the the position of the zombie is if we move would be on the of the from the content of this query which changes but not not in the words of direction so we can distinguish the position of its own the following the vertical direction beyond the optimal cases that's see that is a corner In fact that we can distinguish direct together and only example position of his own so the best candidate to be the futurists of the corner because they have no annotation so we need a corner detection algorithms this article should be should be as skinny body and there is only is shown in the figure in the future I mean by an ugly can we recognize that lines but it will recognize this single
line but bigger as to what the colonists for we need the asking by the solution
is there's scale-invariant feature transform by the law here are presented to the 1 presented the basic steps of so we the press all of a difference of Gaussians operator is applied to the major banks is that the standard deviation the result is a distribution and the extent of the distribution of the scaling by lots different different of Gaussians to detect the corners and edges are not good like every score on the detector is used to live on the edges through obtain a rotation violence and adaptation is assigned to every point then for each key point of our descriptor vector is created the key points on the future of recognized and the key matching is performed through nearest neighbor algorithm between descriptors obviously as the values imagery mission algorithms already implemented and that it does not don't binding let's see an example using SIFT algorithm scaling by form and that I by the them amount of this all we import a major in the class white in the image Commission ugliness are cholera colonists independence we instance data yeah we instantiate perceived algorithm every major recognition are going to immigrants to me as a function of the detector and computer that reads answer key points and descriptors
this function have book for parameters for image information that can be optimized for the specific case see also as a function to represent the points and their orientation on the image do you see from the cholera Quality quotes are the key points to recognize and the they ever annotation
we performed the same process on the imaging in the library we can immediately that we can have several images in our library we matched script so we then it's neighbor the algorithm called then we start the good matches following laws pressure test described in the article Starting from the number of the key points match is that we can understand that you think the major as being nice so we can set up additional on the number of matches of all which we can say that we have recognized the feature the Special these closely related to the algorithm and the whole and to their parameters used for the recognition In this slide I
II show an example of the major recognized and not recognized as we can see that the number of matches significant significantly different for this example I used specifically seek the operator for our application we use we use the surfer I believe that this up approximation of we use it term because it is more proper computationally performance so for a real time recognition is better but the number of documents filed by this algorithm uses immediately lower than the difference between the recognized and not recognize case the fat so it can be very small so so we have to find a strategy to avoid false positives we
develop an algorithm to compute the position of the observer so we have we will have a middle to exclude false positive let's try to compute the position of the several respected image we have
to find the transformation that link the library major and the picture in in the portal is the summation called all modified and the link different projective appropriately projective plane we had 2 cameras a and B looking at the same point B in a plane the projection of being in a and b are respectively the a and B and we can express BA in function of fraction of Pb case and M. M. What is the mother of the and it that it can be expressed through our that the teaser artificial medics and keep that these other solution and the decays mydriasis are becoming thinks experiment so we can compute them can this process is called chemical production and that it is performed through just for me the we take pictures of such as work from different angles and we find the corners of the transport we the aerosol belief of for example and we find a distortion to things 3 lines we can see that in this for the lines of the just a slightly distorted and we have to correct this year this process is already implemented the eating of interval and see crudely functions fine just work on this and anybody Starting from these parameters we can apply formation to be a major now we can see that underlying of which is what are more straight now we
can compute the position of the major of the picture in the image of although we to extract the matching keypoints of days nature and for the library major you can see the function find demography extracted your mother because permission from 2 sets of points and now we can create an array we therefore corners of the picture and then transformed the points that we D. omega phonetics M this is done by the open sea functions that perspective transform and we note that the computed over the years ago that because the rest of the red rectangle is all the pictures in the in the image let's see that we can use these needed to exclude the false positive In this case we have a lot of matches so if we used to think that we have recognized that the picture but the picture is not the same uh if we compute the position this picture positioning that we can know that he is wrong in fact directly that the red rectangle does not fit with the real position of the pitch or so we we have a of what's positive and we we have found that the because of of the of the proposition we tested the imagination
also in the case and we can start with a picture of the cost of buying arc they came from the right a major recognition in In this case it doesn't doesn't work because there are 2 main differences between the images so we the image of
nature works so even in the libraries included also major of the year from the right so we need at least 3 images for in the library from left right and center obviously we can apply these algorithm only the case of front of objects very characteristic as ox or churches therefore 350 cases
we use the imagination in bluegrass application it is arteries guide application that police media contents of based on local for now it was tested in the archaeological area of the the pattern In this
application magic mission is to use the for other parts of the location based on what you're watching and based on the planning of the of the please and the user is used to play at the answer the information the artwork you're watching the thank you for
your attention the and if you have any
questions how do you questions the thank you know I have a question do you do any other transformations to enhance the image quality before you do any processing like for example are it trying to detect if the images blurring or terrorists like sparks from I know the street lights or something the can you repeat again didn't you let's say you take a little picture with the mobile phone and the camera is and you know focused properly I want to take a picture as it is over there do you have other algorithms how to improve the picture so you can run your analysis not we tested on different camera and then every camera as an additional metrics that correctly the destruction of the camera but the he made use it must be the focus on to the blurring gain is not but I have a new word example you had that red square and you could easily see that it was in some cases wrong but how does the computer sees that is from OK 1 moment we these competition we we can compute also the position of the observer and then we have to use the fact that rotation and translation from decommodified sort of in and also to and useful functions 1 is so that can be that problem up well antiques and this that the power coming 25 minutes and that the distorted a major you extract the revision vectors and the translation vectors from these obviously of salts implemented the Rodrigues that you and I believe that the extracted from the addition vectors are addition that and so we can compute the translation in the system reference of the images of the picture and then if the translation is the is use of is wrong and so if the decision is back of the image of for example we can exclude the use of music uh so that in to to or to download the respect of the picture we can exclude that the user is matched thank you more questions the question had the the high is this presentation available somewhere in yes it is uh the link this is the we denote so you can
find online on my linking
pages 1 more question 1 more
question and the associated
library these you only need to recognize that a certain number of features right and how large and is that the library for a real time processing just 10 features hundreds of thousands uh we tested reader that then pictures in the library but we can we can improve this number back to the the time is that the computation time is increases so we can probably lies at the front of the because of In the event more questions and when you do image-recognition that I often see green lines going outside of the red square so basically you find the position of the image on you know based image and why do you assume features extracted outside right there why don't you dream only 2 features so that inside the square I when the future outside the picture we want to compare them because you the people in for example In this example we OK so that we can go to the are example where you which is sort of a graph and some features of young places it is sitting the lines below the red square where you will make the the performance of the data obviously for their position that so the he points outside the images so the wrong key points are uh are less than the set the key points so the defeat exclude them the
probably read it is clear that
of this I have a question about
the bimodality gate where do you do the processing on the class so that you send it somewhere to make sense of what to do moment by that you will suffer in your application for the historical sites you said that you display information depending on where you are in the society and for that you need to figure out where you are so you comparing 2 any image in in your life right the it's locality extinct with the GPS and the but the hero for gps is the order of meters and then we we can find your organization approval of true extract what you're watching in the in the in the so we can preselect the the object that that you're watching it the new the newest thing you're opted to what did you any do the processing that you have in your image right you compare what you see with the library is that you're on the glass not the service I think when you are converting to grayscale exhaustive potentially using additional information a just a flat price cut computation already doing the cold convention already doing any kind of optimization of the process of moving into Christ we use the because it is is more competition the performance and I mean the full the full hearings set algorithm your investing in a color image to gray scale everything at least you suggest a what I thought that was what you do in class II doing just a committee is there any kind of special grayscaling get doing or is it just the conventional agency examples don't have special knowledge justice and my the noblest tensions on the thank you for your attention
Metropolitan area network
Spezialrechner
Offene Menge
Ortsoperator
Elektronischer Programmführer
Software
Open Source
Vorlesung/Konferenz
Mustererkennung
Vektorraum
Extrempunkt
Computeranimation
Invariante
Portscanner
Metropolitan area network
Quader
Spezialrechner
Algorithmus
Datenfeld
Benutzerschnittstellenverwaltungssystem
Evolute
Datenerfassung
Programmbibliothek
Strategisches Spiel
Information
Smartphone
Implementierung
Krümmung
Ortsoperator
Minimierung
Rechteck
Stellenring
Abfrage
Kontextbezogenes System
Computeranimation
Invariante
Richtung
Algorithmus
Vertikale
Keilförmige Anordnung
Flächeninhalt
Flächentheorie
Vorlesung/Konferenz
Wort <Informatik>
Inhalt <Mathematik>
Figurierte Zahl
Gerade
Resultante
Kreisbewegung
Distributionstheorie
Subtraktion
Punkt
Ikosaeder
Computer
Transformation <Mathematik>
Drehung
Gesetz <Physik>
Computeranimation
Spezialrechner
Bildschirmmaske
Algorithmus
Maßerweiterung
Gerade
Distributionstheorie
Lineares Funktional
Zentrische Streckung
Nichtlinearer Operator
Stochastische Abhängigkeit
Vektorraum
Matching
Mustererkennung
Invariante
Extreme programming
Datenfeld
Anpassung <Mathematik>
Schlüsselverwaltung
Standardabweichung
Softwaretest
Parametersystem
Lineares Funktional
Orientierung <Mathematik>
Prozess <Physik>
Punkt
Matching <Graphentheorie>
Zahlenbereich
Mustererkennung
Gesetz <Physik>
Computeranimation
Rechenschieber
Druckverlauf
Algorithmus
Programmbibliothek
Skript <Programm>
Information
Nichtlinearer Operator
Subtraktion
Approximation
Matching <Graphentheorie>
Ortsoperator
Zahlenbereich
Kartesische Koordinaten
Mustererkennung
Term
Computeranimation
Algorithmus
Echtzeitsystem
Strategisches Spiel
Luenberger-Beobachter
Ebene
Kreisbewegung
Subtraktion
Gewichtete Summe
Punkt
Prozess <Physik>
Ortsoperator
Natürliche Zahl
Matrizenrechnung
Rechteck
Transformation <Mathematik>
Computeranimation
Metropolitan area network
Spezialrechner
Arithmetischer Ausdruck
Programmbibliothek
Punkt
Gerade
Normalvektor
Parametersystem
Lineares Funktional
Bruchrechnung
Matching <Graphentheorie>
Winkel
Aussage <Mathematik>
Vorzeichen <Mathematik>
Matching
Biprodukt
Portscanner
Matrizenring
Ebene
Menge
Offene Menge
Parametersystem
Dateiformat
Translation <Mathematik>
Projektive Ebene
Verzerrungstensor
Persönliche Identifikationsnummer
Warping
Neuronales Netz
Subtraktion
Unterring
Natürliche Zahl
Computeranimation
Kreisbogen
Kreisbogen
Objekt <Kategorie>
Spezialrechner
Metropolitan area network
Algorithmus
Rechter Winkel
Programmbibliothek
Vorlesung/Konferenz
Hypermedia
Spezialrechner
Flächeninhalt
Elektronischer Programmführer
Mereologie
Mustersprache
Hypermedia
Automatische Handlungsplanung
Kartesische Koordinaten
Information
URL
Inhalt <Mathematik>
Information
Computeranimation
Prozess <Physik>
Momentenproblem
Ortsoperator
Versionsverwaltung
Drehung
Transformation <Mathematik>
Computer
Kombinatorische Gruppentheorie
Computeranimation
Algorithmus
Gruppe <Mathematik>
Translation <Mathematik>
Luenberger-Beobachter
Vorlesung/Konferenz
Analysis
Leistung <Physik>
Addition
Lineares Funktional
Linienelement
Physikalisches System
Vektorraum
Binder <Informatik>
Fokalpunkt
Entscheidungstheorie
Quadratzahl
Wort <Informatik>
Portscanner
Kreisbewegung
Metropolitan area network
Spezialrechner
Total <Mathematik>
Translation <Mathematik>
Gasströmung
Aliasing
Information
Bitrate
Computeranimation
Homepage
Punkt
Ortsoperator
Graph
Zahlenbereich
Computerunterstütztes Verfahren
Extrempunkt
Ereignishorizont
Quick-Sort
Computeranimation
Vorhersagbarkeit
Metropolitan area network
Echtzeitsystem
Quadratzahl
Programmbibliothek
Vorlesung/Konferenz
Gerade
Managementinformationssystem
Metropolitan area network
Spezialrechner
Elektronischer Fingerabdruck
Computeranimation
Videospiel
Zentrische Streckung
Web Site
Prozess <Physik>
Krümmung
Selbst organisierendes System
Minimierung
Klasse <Mathematik>
Stellenring
Kartesische Koordinaten
Computerunterstütztes Verfahren
Objekt <Kategorie>
Dienst <Informatik>
Verknüpfungsglied
Algorithmus
Rechter Winkel
Programmbibliothek
Meter
Vorlesung/Konferenz
Information
Kantenfärbung
Ordnung <Mathematik>
Schnitt <Graphentheorie>
Unimodale Verteilung

Metadaten

Formale Metadaten

Titel Image recognition and camera positioning with OpenCV. A tourist guide application.
Serientitel EuroPython 2015
Teil 46
Anzahl der Teile 173
Autor Nazzaro, Francesco
Lizenz CC-Namensnennung - keine kommerzielle Nutzung - Weitergabe unter gleichen Bedingungen 3.0 Unported:
Sie dürfen das Werk bzw. den Inhalt zu jedem legalen und nicht-kommerziellen Zweck nutzen, verändern und in unveränderter oder veränderter Form vervielfältigen, verbreiten und öffentlich zugänglich machen, sofern Sie den Namen des Autors/Rechteinhabers in der von ihm festgelegten Weise nennen und das Werk bzw. diesen Inhalt auch in veränderter Form nur unter den Bedingungen dieser Lizenz weitergeben
DOI 10.5446/20172
Herausgeber EuroPython
Erscheinungsjahr 2015
Sprache Englisch
Produktionsort Bilbao, Euskadi, Spain

Inhaltliche Metadaten

Fachgebiet Informatik
Abstract Francesco Nazzaro - Image recognition and camera positioning with OpenCV. A tourist guide application. OpenCV Python bindings provide several ready to use tools for camera calibration, image recognition and camera position estimation. This talk will show how to recognize a picture, from a library of known paintings, and compute the camera position with respect to the recognized picture using OpenCV and numpy. This is applied to a tourist guide application for Google Glass through the recognition of the paintings exposed in the museum.
Schlagwörter EuroPython Conference
EP 2015
EuroPython 2015

Ähnliche Filme

Loading...