The Compressing Sensing Reconstruction Algorithm Based On Backtracking And The Application In Depth Image Processing

With the development of autonomous robots,how to let the robot "understand" the outside world is the content of many scholars.The robot can have contact with the external environment through the visual system.The information processing unit is an essential part of the vision system.It can be realized by hardware when the real-time requirement is high,and can achieve this functionality through the PC unit when the real-time requirement is not high and low power consumption requirement of the circumstances.The speed of transmitting the image information that collected is restricted by the bandwidth.The theory of compressed sensing is a theory of recovering the original signal from a small amount of collected information.In this thesis,an improved compressed sensing reconstruction algorithm is proposed,which is called Backtracking Regularized Stagewise Orthogonal Matching Pursuit algorithm.This algorithm is improved on the basis of the Stagewise Orthogonal Matching Pursuit algorithm,and the optimal atom is selected by regularization operation and backtracking operation.Through the simulation of one-dimensional random signal and two-dimensional images,compared the Stagewise Orthogonal Matching algorithm at the same sampling rate,this algorithm improves the peak signal-to-noise ratio of 20%~40%,and compared with the Orthogonal Matching Pursuit algorithm,it improved by 8%~10%,the reconstruction time is reduced by 70%~80%.The application of BR-StOMP algorithm in depth image processing was studied.Firstly,the edge feature of depth image is analyzed,the depth image filtering preprocessing is performed to observe the filtering effect,and the image edge is detected and the image is blocked,it BR-StOMP algorithm was verified on the depth of image restoration effectiveness for experiments.In the experiment,depth image and color image are acquired with Kinect.In this thesis,the calibration of the camera and the color depth of the camera are calibrated,the radio and tangential distortion are acquired.Finally,through analysis the peak signal to noise ratio and observe the point cloud image,BR-StOMP algorithm can effectively reconstruct the depth image.