Design and Implementation of a 2D MIMO OCC System Based on Deep Learning

Ones Sanjerico Sitanggang, Van Linh Nguyen, Huy Nguyen, Radityo Fajar Pamungkas, Muhammad Miftah Faridh, Yeong Min JangSensors 2023 · DOI

TL;DR

This paper presents a novel Optical Camera Communication (OCC) system designed for high-mobility scenarios, leveraging the YOLOv8 deep learning model for robust LED detection and tracking. The system achieves reliable data transmission even with camera speeds up to 10 m/s, addressing key challenges in current OCC technology.

What Problem Does This Paper Address?

Optical Camera Communication (OCC) is a promising wireless technology offering advantages like high-speed transmission and enhanced security compared to radio frequency. However, its practical implementation faces significant hurdles, particularly in environments with high mobility and varying lighting conditions. These challenges include:

  • Optical Flow Distortion: Rapid camera movement can distort the image, making it difficult to accurately identify and track the light-emitting diodes (LEDs) used for data transmission.
  • Interference and Noise: External light sources like sunlight and streetlights can interfere with the signal, degrading data quality.
  • Real-time Processing Demands: Accurately detecting and decoding OCC signals in real-time requires significant computational power, which can be a bottleneck for embedded systems.

This paper aims to overcome these limitations by developing a robust OCC system that can maintain reliable communication even under challenging mobility and lighting conditions.

Methods

The core of this research lies in the integration of deep learning with established image processing techniques for OCC. The key methods employed are:

  • YOLOv8 for LED Detection and Tracking: The authors utilize the You Only Look Once version 8 (YOLOv8) object detection model. YOLOv8 is chosen for its lightweight architecture, speed, and accuracy, making it suitable for real-time applications and resource-constrained devices. It's employed to precisely identify and track the LEDs that act as data transmitters.
  • Adaptive Thresholding for LED Segmentation: To address issues with varying lighting and noise, the system uses adaptive thresholding for segmenting the detected LEDs. This method is more robust than global thresholding, as it adjusts the threshold locally based on pixel neighborhoods, leading to better segmentation in complex environments.
  • Contour Filtering: After adaptive thresholding, contour detection is used to identify the boundaries of the LEDs. A Gaussian filter is then applied to these contours to refine them, remove noise, and isolate the LED shapes for accurate data extraction.
  • 2D OOK-MIMO Architecture: The system is built upon a 2D On-Off Keying (OOK) Multiple-Input Multiple-Output (MIMO) framework. This architecture uses an LED array as the transmitter, where the presence or absence of light represents binary data (1 or 0). The MIMO aspect likely refers to the use of multiple LEDs for transmitting data, potentially increasing the data rate.

Datasets

The authors collected and annotated a custom dataset for training and validating their YOLOv8 model.

  • Dataset Size: 5000 images were collected.
  • Split: 4000 images were used for training, and 1000 images were reserved for validation.
  • Annotation: The images were annotated using labeling software to identify and mark the LEDs.

Key Insights and Ideas

  • YOLOv8's Suitability for OCC: The study demonstrates that YOLOv8, a state-of-the-art object detection model, is highly effective for the specific task of LED detection and tracking in OCC systems, even in dynamic environments.
  • Adaptive Thresholding's Robustness: The paper highlights the superiority of adaptive thresholding over basic thresholding for LED segmentation in OCC, especially in challenging outdoor conditions with significant noise.
  • Real-time Performance in Mobility: The system's ability to maintain low bit error rates (BER) at high camera speeds (up to 10 m/s) and moderate distances (up to 2m) is a significant achievement, pushing the boundaries of OCC in mobile applications.
  • Lightweight Model for Embedded Systems: The choice of YOLOv8 and the proposed methods are geared towards enabling OCC on resource-constrained embedded devices, a crucial step for widespread adoption.

Key Results

  • High Detection Accuracy: The YOLOv8 algorithm achieved high accuracy, with mAP0.5 and mAP0.5:0.95 values of 0.995 and 0.8604, respectively.
  • Real-time Processing Speed: The system achieved processing speeds up to 1.25 ms, enabling real-time operation.
  • Low Bit Error Rate (BER) in Mobility: The system demonstrated a BER of 10⁻² even at a camera speed of 10 m/s and a distance of 2 meters.
  • Data Transmission Rate: The system is capable of achieving data transmission speeds of up to 3.84 kbps with an 8x8 LED array configuration, which can increase with larger arrays.
  • Performance Comparison: YOLOv8s, particularly with a batch size of 32, outperformed other YOLOv8 variants and YOLOv5 in terms of processing time, weight, and GFLOPs, while maintaining comparable mAP and precision.

Limitations and Future Work

  • Distance Limitations in High Mobility: The paper acknowledges that while the system performs well at close ranges (e.g., 2m) even with high mobility, performance degrades significantly at longer distances (e.g., 8m and 10m) under high mobility conditions. This is identified as a significant challenge for future OCC systems.
  • Future Research Directions:
    • Laser Transmitters: Exploring the use of lasers as transmitters to increase transmission distance.
  • Novel Modulation Schemes: Investigating new modulation techniques to boost data rates.
  • Cloud-Edge Architecture: Developing lightweight OCC approaches integrated with cloud-edge architectures for enhanced processing on resource-constrained devices.
  • Real-World Validation: Conducting further validation of the proposed scheme in actual real-world scenarios.

Summary updated

BibTeX
@article{sitanggang2023design,
  author = {Ones Sanjerico Sitanggang and Van Linh Nguyen and Huy Nguyen and Radityo Fajar Pamungkas and Muhammad Miftah Faridh and Yeong Min Jang},
  title = {{Design and Implementation of a 2D MIMO OCC System Based on Deep Learning}},
  journal = {Sensors},
  year = {2023},
  doi = {10.3390/s23177637},
}
Figure