Adaptive selection of two channels in optical camera communication utilizing deep neural networks

With the increasing popularity of mobile cameras, optical camera communication (OCC) has garnered significant attention. However, in indoor environment, blockages and shadows pose frequent concerns. To address this issue, we developed a dual-channel OCC system utilizing front and rear rolling-shutter cameras, and proposed an adaptive method for selecting the optimal channel through image processing utilizing deep neural networks (DNNs), ensuring seamless switching to the alternative channel in case of blockage, thereby maintaining robust communication performance. After image pre-processing, we adopted YOLOv8n, a detection DNN, to detect the preamble of the data packet in stripe pattern images. This approach exhibited superior performance compared to traditional thresholding-based algorithm. Subsequently, the data packet was recovered in the form of stripe pattern images. To evaluate the quality of each pair of stripe pattern images captured by both channels and automatically select the optimal channel for further decoding, we employed VGG16, a classification DNN. This method demonstrated a higher accuracy and a shorter inference time compared to the eye-diagram algorithm and k-means clustering scheme. Consequently, in the dual-channel system, as long as one of the two channels exhibits a low bit-error rate (BER), the dual-channel system can achieve a low BER, regardless of the communication performance of the other channel.