Learning

Emotion Recognition in Images using Important Feature Extraction

Emotion recognition is a field of research that has a lot of possibilities. For example, emotion recognition makes it possible to recognize the emotions of people to detect possible danger in a video such as from a CCTV. Emotion recognition consists of three steps. First, the face is detected. And then, the features are extracted from the facial images. Finally, the emotion is recognized by classifying the emotion with the extracted features. To recognize emotion, face detection is an important precursor. We used two popular face detection techniques, one is Haar cascades, and the other is LBP features. After detecting face, it is able to get facial landmarks. We used 68 facial landmarks method available in Dlib. It can detect jaw line, eyebrows, eyes, nose, mouth with high accuracy. Using these landmarks, features can be extracted. We reflected the essential features such as shape of eyebrows and mouth, degree of eyes and mouth opening, and ratio from center of face to key landmarks. In this study, we used previous basic techniques of face detection (Haar cascades), then used 68 facial landmarks in Dlib to extract features for emotion recognition, and SVM is used for the learning to recognize emotions.

 

     

    

 


Classifier comparison for bearing failure detection of induction motors using current signal

  Induction motor is widely used in the industry area and the bearing is one of the key mechanical components. The bearing minimizes the friction between the rotating part and stationary part of the rotating machine. It is important to monitor the bearing condition to give a warning before serious failures occur. The fault detection through electrical monitoring has been studied for the last several decades. Although they detect warning signs before serious problems occur, it does not always work when the sampling time is short. This research proposes a learning model for induction motor to diagnose bearing failures which learns features from electrical signatures. This experimental study uses data obtained from 415V, 55KW induction motor and clearance modified plain bearings.