EEG-Based Target Recognition for Multi-camera Surveillance Systems

This paper investigates the integration of brain-computer interfaces (BCIs) for target recognition in surveillance scenarios, utilizing event-related potentials (ERPs) to detect specific stimuli. A novel experimental setup is designed to provoke P300 responses from participants observing surveillance video. EEG data were collected from non-expert and expert subjects. A hierarchical convolutional neural network framework is proposed to effectively manage the varying levels of classification complexity for target, non-target, and background stimuli. The experimental evaluation was conducted in both subject-independent and subject-dependent conditions. In the non-expert dataset, the F1 scores were 0.3843 for subject-independent and 0.4185 for subject-dependent conditions. In the expert dataset, the F1 scores achieved were 0.5807 for subject-independent and 0.6222 for subject-dependent conditions. These performances surpassed the baseline models, demonstrating the feasibility and potential of BCI technology in target recognition tasks. The significant performance difference between non-experts and experts highlights the role of cognitive skills in target recognition tasks.