The increasing risk of drowsiness-related incidents in transportation and daily life highlights the need for an effective, non-intrusive monitoring method. Addressing this, our study introduces a novel drowsiness detection model using electrocardiogram (ECG) data, implemented through a 1D Convolutional Neural Network (CNN) Autoencoder, and further enhanced by incorporating Variational Autoencoder (VAE) techniques. This hybrid model, designed for wearable sensor applications, excels in complex feature extraction from ECG data, crucial for accurate drowsiness detection. The model was evaluated using two-channel ECG signals from the MIT Arrhythmia database, a benchmark for assessing the reliability and accuracy of cardiac monitoring systems. It demonstrates superior performance over traditional methods, including standard 1D CNN and LSTM Autoencoders, particularly in handling noise, artifacts, and the complex temporal dynamics of ECG signals. The efficiency of the model is underscored by its computational speed, with a total training time of just 1.06 minutes, and its effectiveness is evidenced by its lower mean squared error (MSE) and mean absolute error (MAE) compared to other models. This research advances wearable health technology, offering a promising tool for enhancing transportation safety and personal health monitoring through effective, real-time drowsiness detection. It marks a significant step forward in preventing accidents and ensuring well-being through innovative, data-driven solutions.