Distributed video coding for wireless multi-camera networks

Nantheera Anantrasirichai, Dimitris K. Agrafiotis, David Bull2008 · DOI

TL;DR

This paper introduces a novel distributed video coding (DVC) framework for wireless multi-camera surveillance systems. It leverages inter-camera correlations at the decoder to improve side information generation, leading to significant bitrate reductions and better quality compared to existing methods.

What Problem Does This Paper Address?

Low-power video encoders are crucial for wireless surveillance systems where power is limited and network communication between cameras is often undesirable due to strain on resources and increased vulnerability to errors. Traditional video compression methods place a high computational burden on the encoder. Distributed Video Coding (DVC) offers a solution by shifting complexity to the decoder, making it ideal for low-power encoders. However, effectively exploiting the spatial and temporal correlations present in multi-view video sequences at the decoder for DVC remains a challenge. This paper aims to address this by proposing a DVC framework that specifically exploits inter-camera correlations to enhance the quality of the reconstructed video, particularly for surveillance applications.

Methods

The proposed framework splits video frames into Key and Wyner-Ziv (WZ) macroblocks (MBs). Key MBs are intra-coded, while WZ MBs are quantized and their parity bits are sent to the decoder. At the decoder, the side information (SI) for WZ MBs is generated using a combination of temporal and spatial error concealment (TEC/SEC) methods. A key innovation is the use of inter-camera correlations to improve SI generation. Two multi-view coding structures are proposed to exploit these correlations: Structure 1 prioritizes views that can better cover the entire scene, while Structure 2 exploits redundancy between closely located adjacent views.

Enhancements include:

  • Spatio-temporal concealment (STC): Generates SI on a macroblock basis, leveraging both temporal and spatial information.
  • Mode selection: Dynamically switches between TEC and SEC based on the activity within the MB neighborhood.
  • Local correlation noise estimation: Estimates correlation noise on a per-MB basis.
  • Modified B-frame quantization: Increases the quantization step size for WZ MBs in B-frames to further reduce bitrate, leveraging the improved SI quality for these frames.

Datasets

The performance of the proposed codec was evaluated using the "Breakdancers" multi-view sequence at CIF resolution.

Key Insights and Ideas

  • Exploiting inter-camera correlation at the decoder is key: The paper demonstrates that by intelligently using information from neighboring cameras at the decoder, the quality of the side information can be significantly improved, leading to better overall compression performance.
  • Macroblock-level processing is crucial for DVC in multi-view scenarios: Instead of treating entire frames, processing at the MB level allows for more granular exploitation of both temporal and spatial/view correlations, leading to more adaptive and efficient compression.
  • A hybrid approach to SI generation is effective: Combining temporal and spatial error concealment, with a mode selection mechanism, allows the system to adapt to different content characteristics and achieve better SI quality.
  • B-frame optimization can yield significant bitrate savings: By increasing quantization for WZ MBs in B-frames, the paper shows a way to further reduce bitrate without a noticeable subjective quality drop, thanks to the improved SI.

Key Results

The proposed scheme shows considerable performance improvements:

  • Up to 23% bitrate reduction compared to corresponding frame-based schemes.
  • Improvements of up to 1.4 dB in PSNR relative to an equivalent intra-view codec.
  • The two proposed multi-view coding structures (Str1 and Str2) showed similar results, suggesting further investigation with more diverse multi-view sequences might be beneficial.

Limitations and Future Work

  • Limited testing: The authors suggest that more multi-view test sequences should be used, as the performance might be affected by the baseline geometry of the cameras (wide vs. narrow).
  • Future work: The paper implicitly suggests exploring more advanced multi-view correlation exploitation techniques and potentially more sophisticated SI generation and correlation noise estimation methods.

Summary updated

Abstract

In this paper, we present a novel distributed multiview video compression framework for surveillance purposes that employs a concealment based approach for the generation of the side information through the use of hybrid Key/Wyner- Ziv frames. Two multi-view coding structures are proposed in order to better exploit the inter-camera correlation present at the decoder. We additionally introduce a number of enhancements including use of spatio-temporal concealment for generating the side information on a MB basis, mode selection for switching between the two concealment approaches and for deciding how the correlation noise is estimated, local (MB wise) correlation noise estimation and modified B frame quantisation. The results presented indicate considerable improvement (up to 23%) compared to corresponding frame based schemes and improvements of up to 1.4 dB relative to an equivalent intra-view codec.
BibTeX
@misc{anantrasirichai2008distributed,
  author = {Nantheera Anantrasirichai and Dimitris K. Agrafiotis and David Bull},
  title = {{Distributed video coding for wireless multi-camera networks}},
  year = {2008},
  doi = {10.1049/cp:20080285},
}
Figure