CNN-Based Human Detection and Identification in Indoor Optical Camera Communication Systems Using a Wearable LED Strip

Abstract

In this paper, we present a proof of concept for an indoor optical camera communication (OCC) system utilising a deep learning network to detect and identify humans wearing light-emitting diode (LED) strips. Specifically, we propose using the You Only Look Once (YOLO) version 8 object detection algorithm, which is built on convolutional neural networks (CNNs), to identify wearable LED transmitters in challenging scenarios such as low visibility, mobility and multiple users, followed by image processing to effectively decode the transmitted data. The red-green-blue (RGB) LED strip's colours (red, green, blue and white) serve as indicators of the user's status. By combining communication and monitoring functionalities, the LEDs facilitate not only the transmission of user data but also accurate detection, tracking and identification within the environment. This demonstrates the feasibility of utilising widely available devices like LED strips and cameras, commonly found in many buildings, with potential applications in high-risk environments where monitoring individuals' physical conditions is crucial. The obtained results indicate our system's effectiveness, as it achieved up to 100% success of reception (SoR) in a static experimental setup, 96.2% in a walking experimental setup with one user and showed no effectiveness with two users.