Drowsiness reduces reaction time, leading to potentially fatal accidents. Most existing research focuses on only one symptom of drowsiness, which often results in false alarms. This paper introduces a novel approach for real-time drowsiness detection. The proposed method employs four deep learning architectures based on convolutional neural networks: AlexNet for environmental feature extraction, ResNet50V2 for hand gesture recognition, VGG-FaceNet for facial feature extraction, and FlowImageNet for behavioral feature analysis. To improve the voting system, the paper suggests using a layer with adjustable learning weights to reduce false results and enhance accuracy. Testing the static method on the NTHUDDD dataset and a custom dataset demonstrates that the proposed approach achieves higher accuracy (97.25% and 96.75% respectively) compared to existing methods.