High-Speed Extraction of Regions of Interest in Optical Camera Communication Enabled by Grid Virtual Division

Xin Hu, Pinpin Zhang, Yimao Sun, Xiong Deng, Yanbing Yang, Liangyin Chen

TL;DR

This paper introduces a novel "Grid Virtual Division" (GVD) scheme to significantly speed up the extraction of Regions of Interest (RoI) in Optical Camera Communication (OCC). By virtually dividing images into blocks and sampling pixels, their method reduces decoding delay by up to 70% compared to traditional Gaussian blur techniques, paving the way for higher communication capacities in OCC systems.

What Problem Does This Paper Address?

Optical Camera Communication (OCC) leverages everyday LED lights and smartphone cameras for wireless data transmission, offering a promising alternative to traditional radio frequency methods. However, a major bottleneck in OCC systems is the time-consuming process of extracting the Region of Interest (RoI) – the area in the camera's view that contains the transmitted light signals. Existing methods, often relying on Gaussian blur and complex image processing, are too slow, limiting the overall communication speed and capacity. This paper aims to tackle this performance bottleneck by developing a faster and more efficient RoI extraction method.

Methods

The core innovation is the Grid Virtual Division (GVD) scheme. Instead of processing every pixel in the image, GVD works as follows:

  1. Image Chunking: The received image is virtually divided into a grid of smaller blocks (e.g., 6x6). This division is conceptual and doesn't involve creating separate image subgraphs, thus saving computational resources.
  2. Pixel Sampling: Within each virtual block, the algorithm randomly samples a few pixels from predefined areas (left, center, right) of the middle row. This is based on the assumption that areas with high grayscale values (indicating the LED signal) are likely to be concentrated and will be detected by these random samples.
  3. Thresholding: The sampled pixel grayscale values are compared against a pre-defined threshold. If a sampled pixel exceeds this threshold, the block is marked as potentially containing an RoI.
  4. Multi-block Stitching: Marked blocks are then analyzed to determine their spatial relationships (adjacent blocks). If adjacent blocks are identified, they are "stitched" together to form a more complete RoI. This step accounts for the fact that an RoI might span across multiple virtual blocks.
  5. Demodulation and Decoding: Once the RoI is identified and stitched, a single row or column of pixels from within this region is sampled to form a grayscale sequence, which is then used for demodulation and decoding of the transmitted data.

This approach avoids the computationally intensive Gaussian blur and the need to traverse every pixel, significantly reducing processing time.

Datasets

Not discussed in the paper. The experiments were conducted using a custom-built OCC system with commercially available hardware.

Key Insights and Ideas

  • RoI extraction is a critical bottleneck: The paper highlights that despite advances in modulation and other aspects of OCC, the RoI extraction step remains a significant performance limiter.
  • Random sampling can be effective: The GVD scheme demonstrates that strategically sampling pixels, rather than exhaustively processing the entire image, can efficiently locate the RoI.
  • Virtual division is key to efficiency: By creating a virtual grid, the algorithm reduces the search space for each block without incurring the overhead of actual image segmentation.
  • Grayscale thresholding is sufficient for OCC: For the specific context of OCC, where bright LED signals create distinct high-grayscale areas, a simple thresholding mechanism is effective for initial RoI identification.
  • Stitching is necessary for completeness: The multi-block stitching step is crucial for ensuring that the entire RoI is captured, especially when the LED signal spans across multiple virtual grid cells.

Key Results

The paper presents compelling experimental results demonstrating the effectiveness of the GVD scheme:

  • Decoding Delay Reduction: The proposed GVD scheme achieved a 70% lower decoding delay compared to the traditional Gaussian blur method when tested on an iPhone 8 Plus at a transmission frequency of 5 kHz.
  • Processing Time: The GVD scheme consistently outperformed the baseline in terms of processing time across various transmission frequencies (4, 5, and 6 kbps), with median times significantly lower (e.g., 0.33 ms vs. 1.71 ms at 4 kbps).
  • Impact of LED Shape/Size: The GVD scheme showed superior performance across different LED shapes and sizes, with median processing times for round LEDs being around 0.23-0.28 ms compared to 1.26-1.28 ms for the baseline. For square LEDs, it was around 0.41-0.44 ms versus 0.99-1.99 ms for the baseline.
  • Impact of LED Number: While the processing time increased slightly with more LEDs (due to stitching), the GVD scheme still maintained significantly lower processing times than the baseline.
  • Bit Error Rate (BER): The BER for the GVD scheme remained within reasonable limits, generally below the FEC threshold of 0.38% at up to 5 kbps, demonstrating its feasibility and effectiveness alongside speed improvements.

Limitations and Future Work

  • Threshold Optimization: The paper mentions setting a grayscale threshold empirically. Further research could explore adaptive thresholding methods that automatically adjust based on varying lighting conditions.
  • Complexity of Stitching: While the stitching step is necessary, its complexity could be further optimized.
  • Impact of Environmental Factors: The experiments were conducted in a controlled environment. Future work could investigate the performance under more challenging conditions, such as varying ambient light or motion blur.
  • Deep Learning Comparison: The paper briefly mentions deep learning methods like YOLO but omits their results due to their higher processing times compared to the baseline Gaussian blur. Future work could explore more lightweight or optimized deep learning models for RoI extraction in OCC.

Summary updated

Abstract

Optical camera communication (OCC), enabled by light-emitting diodes (LEDs) and embedded cameras on smartphones, has drawn considerable attention thanks to the pervasive adoption of LED lighting and mobile devices. However, most existing studies do not consider the performance bottleneck of Region of Interest (RoI) extraction during decoding, making it challenging to improve communication capacity further. To this end, we propose a fast grid virtual division scheme based on pixel grayscale values, which extracts RoI quickly without sacrificing computational complexity, thereby reducing the decoding delay and improving the communication capacity of OCC. Essentially, the proposed scheme uses a grid division strategy to divide the received image into blocks and randomly sample several pixels within different blocks to quickly locate the RoI with high grayscale values in the original image. By implementing the lightweight RoI extraction algorithm, we experimentally verify its effectiveness in reducing decoding latency, demonstrating its superior performance in terms of communication capacity. The experimental results clearly show that the decoding delay of the proposed scheme is 70% lower than that provided by the Gaussian blur scheme for the iPhone receiver at a transmission frequency of 5 kHz.
BibTeX
@inproceedings{hu2022high,
  author = {Xin Hu and Pinpin Zhang and Yimao Sun and Xiong Deng and Yanbing Yang and Liangyin Chen},
  title = {{High-Speed Extraction of Regions of Interest in Optical Camera Communication Enabled by Grid Virtual Division}},
  booktitle = {Sensors},
  year = {2022},
  url = {https://openreview.net/forum?id=3XZwHiwTes},
}
Figure