A novel SSVEP-BCI approach combining visual detection and tracking for dynamic target selection

This paper presents a brain-computer interface (BCI) approach for dynamic pedestrian selection. The experimental paradigm is based on steady-state visual evoked potential (SSVEP) and visual detection and tracking. In this study, we collected a few videos of driving environment containing multiple pedestrians, and used the object detection algorithm named YOLOv4-tiny and the multi-object tracking algorithm named Deep-SORT to detect and track pedestrians. In the experiment, these videos are presented on the screen, with flicker stimuli at different fixed frequency superimposed on pedestrian targets. Then subjects selected pedestrian targets according to the prompts. Six subjects (five males, one female) participated in the experiment. Their average response time of SSVEP in offline experiments is 1.98s, in online experiments, the average accuracy is 92.5% and the average ITR is 46.81bits/min. The results show the feasibility and effectiveness of using SSVEPBCI to select dynamic targets in real environments. Object detection and tracking algorithms can detect and track targets in other categories, therefore, this paradigm can also be applied in other scenarios.