Research On0-1Measurement Matrix Based Compressed Sensing Technology Of Visiual Images

Embedded vision is a key issue for mobile robot. It is very important for space adventure, ocean exploration, operation in dangerous working environment, and also for autonomous robotics. However, the image data is large, which usually will seriously affect the system processing speed, the storage performance, and the followed data transmition. The compressed sensing technology appeared in21st century provides a novel methodology for the problem. Centering around the application of the compressive sensing technologies in embedded environmental vision, correlative research has carried out in the paper. The main job is as follows:(1) Sparsity results of environment images are analized based on wavelet sparsity and bandelet sparsity. The effect, stability and advantages are also given both in the higher compression ratio and the lower for the both sparsity methods. For the given256X256environment lawn texture images, reconstruction peak signal to noise ratio (PSNR) is similar for bandelet sparsity and sym8wavelet sparsity (Bandelet sparsity method is slightly better than sym8wavelet sparsity method.), when the measurement value M>96; the PSNR values are all larger than30db, while M=160. unstability of reconstruction is appeared when M≤80for wavelet sparsity method (when M=96,the unstable state appears occasionally.); however, the bandelet sparsity method is relatively stable. In lower measurement value, the bandelet sparsity method is easy to achieve a good reconstruction result.(2) Basic principle of hardware implementation for wavelet sparsity and reconstruction result based on Matlab simulation environment have been give in the paper. It shows that the compressed sensing data reconstruction is practical at the computer terminal for the embedded environmental vision and the compressive sensing method can be used in robotic embedded vision.(3) A novel pseudo-random sequence based0-1measurement matrix is designed, and based on the matrix, the CS reconstruction simulation results are acheived. The result of256x256images shows:the images sparsed by orthogonal wavelet, can be compressed measured by pseudo-random sequence based0-1measurement matrix and effectively reconstructed. To lawn images, the reconstruction PSNR is about34db when M=176, which can be satisfied for usual monitoring.(4) Based on the design of vectorized random0-1matrix, a further design of simple deterministic0-1measurement matrix is carried out and the relevant analysis is completed. The analysis includes reconstruction effect, performance, and the computaion efficiency of the traditional Gaussian measurement matrix(GMM), pseudo-random sequence based0-1measurement matrix(PSMM), and simple deterministic0-1measurement matrix(SDMM). The result shows the SDMM based reconstruction effect is better, the computation efficiency is high and it is suitable for hardware implementation.