Özet:

Abstract The most difficult research problem is ensuring the safety of women and children. Multimodal emotion recognition is a difficult task. One of the most important and widely used research domains in HCI is multimodal data, which includes audio, video, text, facial expression, body motions, bio-signals, and physiological data. This data is used to forecast the safety of women and children. Rigid research has been proposed in this context. To create the best multimodal model for emotion recognition combining picture, text, audio, and video modalities, a novel deep learning model is developed, and a thorough analysis of data, feature, and model-level fusion is undertaken. Separate innovative feature extractor networks are suggested specifically for picture, text, audio, and video data. Then, at the model level, an ideal multimodal emotion identification model is developed by combining information from images, text, voice, and video. Three benchmark multimodal datasets, IEMOCAP, Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS), and Surrey Audio-Visual Expressed Emotion (SAVEE), are used to evaluate the performances of the suggested models. On the IEMOCAP, SAVEE, and RAVDESS datasets, the suggested models obtain high predicted accuracies of 96%, 97%, and 97%, respectively. By contrasting the outcomes with those of the current emotion recognition models, the models' efficacy and optimality are also confirmed. Women and Children Safety Prediction employs multimodal Enhanced emotion recognition.

Anahtar Kelimeler: