| ABSTRACT |
Researchers in neuroscience, computer science, psychology, linguistics are more focused on automated analysis of human facial expressions to interact with computers or robots. The communication between computer and humans are more natural when the machines have an ability to identify and act in response to human non-verbal communication such as emotions. Although a number of approaches have been proposed to distinguish human emotions based on speech or facial expressions, comparatively limited work has been done in real-time recognition system. The limitations of earlier works are: (i) less number of facial expressions are considered for developing emotion recognition system (ii) most of the works have used physical markers (intrusive) for detecting the emotions (iii) the effect of lighting and background changes on facial images significantly affect the decision making ability of the system (iv) very few research works are focused on developing real time systems. In order to elevate the limitations of earlier studies, this product is aim to work on detecting five facial expressions (normal, happiness, sadness, fear, and surprise) using 10 virtual and dynamic markers (non-intrusive) under various lighting and background conditions. All the markers are placed on the human face virtually through proposed marker placement algorithm by focusing into eye and face detection. Furthermore, the all the markers are tracked continuously in online through the help of Optical Flow Algorithm (OFA) in real time scenario and detecting the emotions through facial expressions using rules based system. The output of the system was interfaced with home applications such as light and fan. The emotional state change of the user is used to control the home appliances (Fan and Light) to ¿ON¿ and ¿OFF¿. A circuit was designed and implemented to indicate the connection of microcontroller with the system and home appliances. The end result of this product indicate that, the proposed system is effectively works on various lighting conditions, decision output of this system does not affected by any background changes, and a maximum emotion classification rate of 92.3% is achieved on detecting six different facial expressions.
|
