A MULTIMODAL DRIVER ANGER RECOGNITION METHOD BASED ON CONTEXT AWARENESS

Abstract

In modern traffic situations, driver aggression presents a significant risk to road safety, requiring specific real-time emotion recognition technologies. A unique Context-aware Multi-modal driver Emotion identification (CA-MDER) method is supplied to efficaciously address great issues in multi-modal emotion identity, such as individual variability and contextual sensitivity in emotional expression underneath various driving eventualities. This approach combines facial expressions, verbal indicators, and using status signals to facilitate thorough rage identification. It utilizes an attention Mechanism-enhanced Depthwise Separable Convolutional Neural network (AM-DSCNN) for effective feature extraction, together with improved support Vector Machines (SVM) and Random forest (RF) classifiers for reliable classification. A Contextaware Reinforcement learning (CA-RL) paradigm is utilized for adaptive weight allocation, facilitating dynamic decision-level integration across modalities according to contextual importance. Furthermore, CNN-based transfer learning methodologies are hired to enhance model generalization and classification precision. The system demonstrates incredible performance, with an overall accuracy of 91.68% and an F1 score of 90.37%, indicating its proficiency in accurately identifying and categorizing angry emotions across various data modalities. The implementation has full-size promise for improving using safety by facilitating accurate emotion identification via sophisticated deep learning and adaptive fusion techniques.