Change search
ReferencesLink to record
Permanent link

Direct link
Känsloigenkänning i form av ansiktsuttryck med Kinect
KTH, School of Computer Science and Communication (CSC).
KTH, School of Computer Science and Communication (CSC).
2014 (Swedish)Independent thesis Basic level (degree of Bachelor), 10 credits / 15 HE creditsStudent thesis
Abstract [en]

Facial expressions are a part of our body language that helps us to clarify the verbal communication between humans. We use our facial expressions every day, both consciously and unconsciously, to express emotions and attitudes depending on the situation. The hypothesis of the study is: Given a facial expression, how well can Microsoft Kinect, as an input method, determine a person’s feelings with the two algorithms Naive Bayes and Sequential Minimal Optimization? The feelings are limited to happy, sad, surprised and disgusted. With the help of Kinect, a person’s facial data, both coordinates of the face and parameterized data, were saved and used for machine learning. Two field studies were conducted, where 30 respectively 31 people attended and were instructed to simulate every facial expression mentioned above. In the first field study, coordinates were saved of specific parts of the face, while in the other field study; parameterized data of the face was saved. All data was sent into the machine learning software Weka and was processed with the two algorithms Naive Bayes and Sequential Minimal Optimazation. The best result was given by field study 2, were both Naive Bayes algorithm and Sequential Minimal Optimization gave a success rate of 56,45%. The conclusion of this report is that there is evidence supporting the hypothesis that Kinect, with the help of both algorithms, do recognize facial expressions to a certain extent. However, for the study outcome to be successful a larger sample of data is required for the learning process than what is used in this reports survey.

Abstract [sv]

Ansiktsuttryck är en del av vårt kroppsspråk som hjälper oss att förtydliga den verbala kommunikationen mellan oss människor. Vi använder ansiktsuttryck varje dag, medvetet eller omedvetet, för att uttrycka känslor och attityder till de situationer vi befinner oss i. Frågeställningen i denna rapport är: Givet ett ansiktsuttryck, hur bra kan Microsoft Kinect som inmatningsmetod avgöra en persons känslor med algoritmerna Naive Bayes och Sequential Minimal Optimization? Känslorna avgränsas till glad, ledsen, förvånad och äcklad. Med hjälp av Kinect kan en persons ansiktsdata, både koordinater på ansiktet och parametriserad data, sparas ner och användas för maskininlärning. Två fältstudier gjordes, där 30 respektive 31 personer deltog och fick i uppdrag att simulera varje ansiktsuttryck. I första fältstudien sparades koordinater på vissa specifika delar av ansiktet medan i andra fältstudien sparades parametriserad data av ansiktet. All data skickades in i maskininlärningsprogrammet Weka och bearbetades med algoritmerna Naive Bayes och Sequential Minimal Optimazation. Det bästa resultatet gavs av fältstudie 2, där både algoritmerna Naive Bayes och Sequential Minimal Optimization, gav en träffsäkerhet på 56,45 %. Slutsatsen för rapporten är att det finns en möjlighet för Kinect att med dessa algoritmer känna igen ansiktsuttryck, dock krävs förmodligen en mycket större mängd data för inlärningsprocessen än vad som används i denna rapports undersökning.

Place, publisher, year, edition, pages
National Category
Computer Science
URN: urn:nbn:se:kth:diva-157634OAI: diva2:770756
Available from: 2014-12-12 Created: 2014-12-11 Last updated: 2015-08-27Bibliographically approved

Open Access in DiVA

fulltext(1047 kB)87 downloads
File information
File name FULLTEXT01.pdfFile size 1047 kBChecksum SHA-512
Type fulltextMimetype application/pdf

By organisation
School of Computer Science and Communication (CSC)
Computer Science

Search outside of DiVA

GoogleGoogle Scholar
Total: 87 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

Total: 198 hits
ReferencesLink to record
Permanent link

Direct link