Huge video data has posed great challenges on computing power and storage space,\ntriggering the emergence of distributed compressive video sensing (DCVS). Hardware-friendly\ncharacteristics of this technique have consolidated its position as one of the most powerful architectures\nin source-limited scenarios, namely, wireless video sensor networks (WVSNs). Recently, deep\nconvolutional neural networks (DCNNs) are successfully applied in DCVS because traditional\noptimization-based methods are computationally elaborate and hard to meet the requirements of\nreal-time applications. In this paper, we propose a joint samplingâ??reconstruction framework for DCVS,\nnamed â??JsrNetâ?. JsrNet utilizes the whole group of frames as the reference to reconstruct each frame,\nregardless of key frames and non-key frames, while the existing frameworks only utilize key frames\nas the reference to reconstruct non-key frames. Moreover, different from the existing frameworks\nwhich only focus on exploiting complementary information between frames in joint reconstruction,\nJsrNet also applies this conception in joint sampling by adopting learnable convolutions to sample\nmultiple frames jointly and simultaneously in an encoder. JsrNet fully exploits spatialâ??temporal\ncorrelation in both sampling and reconstruction, and achieves a competitive performance in both\nthe quality of reconstruction and computational complexity, making it a promising candidate in\nsource-limited, real-time scenarios.
Loading....