DEEP LEARNING BASED REAL-TIME FACIAL EMOTIONAL DETECTION AND RECOGNITION

Main Advisor: Wegari Dibaba (Assistant prof.)<br> Artificial Intelligence (AI) is increasingly being incorporated into commonplace applications in the present era. The ability of a computer system to recognize emotions on human faces in real time is one of the most intriguing uses of artificial intelligence. This system is capable of identifying a wide range of emotions, including anger, sadness, and happiness. The process of identifying facial expressions and deciphering the emotion they represent is known as facial emotion recognition. It is an essential component of many applications, including: Surveillance systems for identifying suspicious activity; interactive games and entertainment to improve user experiences; and customer feedback analysis for user engagement. After training the model, we can use it to recognize emotions in real time by integrating it with a live webcam stream. We used Open CV for video capture and face detection. It takes an image or video as input, then detects and recognizes emotions available on that image by putting a bounding box and naming that material. Image data from 7 classes have been collected in image and video format from three person. In order to train our model, preprocessing was carried out, including frame extraction, data resizing, labeling, and splitting using Rob flow, as well as the conversion of images into YOLO model format. Finally, based on the outcomes of our experiment, we can identify and categorize emotions with medium-sized data sets in a timely and accurate manner. The yolov9 algorithm has a high prospect of successfully detecting the emotion in real-time, according to the promising value and forecast results for the image, webcam, and video file. We trained and tested the model using the emotion dataset. The YOLOv9 model achieved a precision of 94.1%, recall of 93.7%, mAP of 94.9% at 0.5 IoU, and a score of 80.4% at 0.5:0.95 IoU, which is suitable for real-time emotion detection.