Random Access Vision-based frame-by-frame gaze direction control driven by object detection

Random Access Vision (RAV) enables the selection of arbitrary gaze directions on a frame-by-frame basis by integrating a high-frequency resonant mirror with a lock-in pixel image sensor capable of multiple exposures within a single frame. This paper presents an evaluation of RAV for object detection tasks, including high-speed simple object detection and simultaneous detection of multiple complex objects. Image binarization is employed to detect rapidly blinking LEDs, and gaze direction is dynamically adjusted frame by frame based on target coordinates. Furthermore, the YOLO algorithm is utilized to detect three distinct objects simultaneously, leveraging the image sensor’s ability to capture three images per frame, each corresponding to a specific target. In the experiment, 1,000 consecutive frames are randomly recorded, and the detection rates for the three targets are 99.5%, 100.0%, and 97.1%, respectively. The average detection time is 4.660 ms, meeting the standard for real-time detection. Experimental results demonstrate stable and continuous tracking of both high-speed simple targets and multiple complex objects, confirming the effectiveness and robustness of the proposed frame-by-frame gaze control method.