A deep learning approach for non-line of sight data in optical camera communications

In the field of optical camera communication (OCC), extracting data accurately from non-line-of-sight (NLOS) situations is challenging due to reflections and bad lighting. To overcome this challenging situation, with the help of deep learning, images that contain data were segmented at the bit level. A customized U-Net centric convolutional neural network was trained with different data that included bad lighting and different reflections. A preliminary analysis of more extensive academic research was completed in order to fully analyze the challenges of OCC. After this, there was discussion about the difficulties that deep learning in the field of OCC had to overcome. This study was conducted since there was not sufficient knowledge in the literature on how to increase the number of bits decoded in the NLOS OCC situation with deep learning segmentation. A large data collection made up of images from different NLOS OCC scenarios was developed. The ability of the model to solve problems and make generalizations was improved with the inclusion of this large data collection. The findings demonstrated that, in this particular situation, the deep learning model performed better than the conventional thresholding technique.