Multimodal emotion recognition

• Autor(a) principal: Silva, Pedro Coelho Grácio da
• Data de Publicação: 2015
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10071/13798
• topic: Emotions; Automatic recognition; 3D; Camera; Multimodal; Audio; Video; Depth; Engenharia informática; Interação homem-máquina; Emoção; 3D; Reconhecimento biométrico; Tecnologia audiovisual

Reading emotions from facial expression and speech is a milestone in Human-Computer Interaction. Recent sensing technologies, namely the Microsoft Kinect Sensor, provide basic input modalities data, such as RGB imaging, depth imaging and speech, that can be used in Emotion Recognition. Moreover Kinect can track a face in real time and present the face fiducial points, as well as 6 basic Action Units (AUs). In this work we explore this information by gathering a new and exclusive dataset. This is a new opportunity for the academic community as well to the progress of the emotion recognition problem. The database includes RGB, depth, audio, fiducial points and AUs for 18 volunteers for 7 emotions. We then present automatic emotion classification results on this dataset by employing k-Nearest Neighbor, Support Vector Machines and Neural Networks classifiers, with unimodal and multimodal approaches. Our conclusions show that multimodal approaches can attain better results.