Research On Compressed Sensing Based Video Codec Technology In WMSN

Due to the ability of rich sensory information and strong extensibility, WirelessMultimedia Sensor Networks (WMSN) has been widely used in many fields. However,multimedia video nodes in WMSN have some limitations on energy and computing power,huge amounts of data processing and transmission would consume a large amount ofcomputing resources and energy, which will accelerate the death rate of video node, and theoverall life cycle of WMSN will be released. Therefore, it is essential to study a videocompression algorithm with low complexity for WMSN to decrease the data transmission,which will prolong the life-cycle of network and consequently guarantee the reliability andvalidity of WMSN monitoring.Recently, the CS (Compressed Sensing) has improved new solution for the videocompression, which has several advantages such as low encoding complexity, competitivecompression performance. Based on CS, DCS (Distributed Compressed Sensing) hasdeveloped with the ability to reconstruct multiple data simultaneously by effectively utilizingspatial-temporal correlation among signals. Therefore, in the light of CS theory and DCStheory, based on the improvement of Compressed Video Sensing(CVS) and DistributedCompressive Video Sensing (DCVS), two kinds of video sensing coding algorithms forWMSN were proposed on account of the video features in WMSN.On the basis of compressed video sensing(CVS), a multi-mode compressedsensing-based video codec method in WMSN was presented by analyzing the characteristicsof video. The proposed algorithm can adaptively group on video according to the localcorrelation of video. For key frame, CS method was used to achieve frame-based randomprojection at the coding end, and at the decoding end, completing the reconstruction processvia the Orthogonal Matching Pursuit (OMP) algorithm For non-key frame inside the group,first the image blocks can be divided into a variety of patterns after mode decision on theencoding end, then different ways of sparse will be adopted in line with the encode mode ofimage blocks. At the decoding end, non-key frames will be reconstructed by integrating thesparse way of image blocks with local edge information. Experiments show that the proposedmethod can effectively reduce the video data transmission and guarantee the reconstructionquality simultaneously.On the basis of distributed compressed video sensing(DCVS)，a DCVS algorithm basedon clustering dictionary learning was put forward. Similar to the method above, we adoptedthe same processing method to key frame. For the purpose of dealing with non-key frame,according to the clustering dictionary learning, the reconstructed image blocks in rapidly changing area and graded collection area will be obtained via blocking and clustering, thenmore precise clustering samples will be achieved after using K-means algorithm. Second,training two clustering sample collections to get two clustering dictionaries. Finally,non-critical frames take use of CS measurements and suitable clustering dictionary toreconstruct at the decoding end. The experimental results show that the reconstructionperformance of the method is better and encoding computational complexity is lower underthe same sampling rate.