The Research Of WMSN Video Reconstruction Algorithm Based On Compressed Sensing

Wireless Multimedia Sensor Networks(WMSN) is a comprehensive network that collects sensors, wireless communication and embedded information. Due to the existed intuitive form data, rich video and other multimedia information, WMSN is widely applied in many fields, such as video surveillance, intelligent security and so on. However, video, as an important data form in WMSN, consumes a lot of computation resource. If taking use of the conventional Nyquist sampling theorem, it will pose a great challenge to the computing power, storage capacity and energy consumption of the WMSN video node, which will seriously affect the network performance and life cycle of WMSN.The recent development of compressed sensing(CS) theory provides a new way for mass data compression. According to CS theory, the traditional limits of the Nyquist sampling theorem can be surpassed by taking advantage of redundancy in the process of sampling, which greatly simplifies the processing procedure of signal compressive sampling. Therefore, it will effectively ease the data processing pressure of resource-constrained WMSN video when utilizing the CS theory into the data compression processing of WMSN video. In this paper, CS-based WMSN video reconstruction algorithm based on CS theory and WMSN video surveillance scenarios was studied, the main research work as follows:On the basis of video compression sensing data acquisition model, and for the special requirements of WMSN video surveillance scenarios, a video reconstruction method based on modified Gradient Projection for Sparse Reconstruction(GPSR) was proposed to ensure a higher real-time performance while improving the quality of the final video. Firstly, preview video can be quickly reconstructed by constructing an appropriate observation matrix. And then, optical flow method was used to extract motion vector in the preview video. Consequently, high-resolution video can be reconstructed by applying Modified GPSR with the motion vectors as constraints. Experimental results show that the proposed algorithm can quickly obtain the preview video, and it achieves a better final performance than GPSR in terms of video subjective quality and objective evaluation index.To further enhance the real-time and effect of reconstruction algorithm for WMSN video surveillance, WMSN video reconstruction algorithm based modified Iterative Shrinkage Thresholding(IST) was presented based on the idea of two-stage reconstruction. The IST data fidelity term was rebuilt by using the motion vector between video frames, parameter was updated through shrinking regularization iterative threshold. Experimental results show that this method can achieve a quick preview WMSN video and meet real-time requirements of video surveillance. In the term of final reconstruction effect, the proposed algorithm has been greatly improved in both visual effect and objective evaluation when compared with the IST algorithm, which makes it preferably meet the quality requirements of video surveillance WMSN.