Object Detection in Surveillance Video using EEG with Domain Adaptation Approach

Deep learning has become an integral technique in Brain-Computer Interface (BCI) research, especially in the area of rapid serial visual presentation, due to its proficiency in interpreting complex electroencephalography (EEG) data patterns. However, deep learning's full potential in BCIs is somewhat limited by issues like data scarcity and notable variabilities both within and between subjects. To address these challenges, this study introduces a deep learning training approach that incorporates an overlapping sliding window technique. Following this, we develop a deep network that utilizes domain adaptation, integrating samples from various subjects to enhance object detection performance in surveillance paradigms. Our findings reveal that this overlapping sliding window method surpasses traditional trial-based methods. Additionally, we note a performance improvement when domain adaptation techniques are employed.