This paper presents an innovative approach for
optical camera communication (OCC) using two-dimensional matrix LED transmitter where symbols are detected and classified using a lightweight YOLO object detection algorithm. The data-sets built for this study can have any number of symbols (not necessarily a power of two), because we are considering a new differential encoding scheme where data is embedded in symbol transitions. Two different data-sets were produced: one with symbols having a matrix identifier and orientation indicator, and the other lacking these elements. We investigated the performance of YOLOx-S on both data-sets under a variety of conditions, including variable distances, angles, LED matrix brightness, camera exposure time, and lighting environments.
The average performance metrics were evaluated for each dataset during training, showing better results with the first data-set. We found that the presence of a matrix identifier significantly improved the detection precision. Furthermore, we examined the
effect of LED matrix reflections on object detection accuracy. Overall, this research presents the feasibility of implementing an OCC system with a computer vision-based receiver, establishing a foundation for the development of more complex communication
systems.