Lens Imaging Geometric Distortion Correction Study Based On Compress Sensing

Along with the computer vision technology is widely used in modern industry. Optical image detection based on computer vision technology for its advantages of high efficiency, high precision and intelligent has gradually replace the traditional detection techniques. But the optical imaging system generated image also inevitable drawbacks: optical distortion, due to the installation of the optical imaging system, the lens processing factors such as object after optical system imaging is the phenomenon of distortion and deflection or tilted. As the requirement of accuracy is increasing, the imaging system of geometric distortion of lens imaging acquisition image distortion problem have not allow to ignore. For high precision optical image measurement system, lens imaging geometric distortion caused by the measurement error is an important index for measuring system. Due to the geometric distortion of lens imaging is nonlinear distortion, so it is difficult to find a general calibration model to meet the testing requirements of high accuracy of the size.Compress sensing once got the extensively interest by the scholars in various fields and already have been applied in many fields. The basic idea of compress sensing is that if a sparse signal or the signal is sparse in a transform domain, you can recover original signal through the signal of a small amount of samples values with high-precision. Revelation of the compress sensing, Considering that we can consider discrete sampling point error value of lens geometric distortion are studied in this paper as sparse signals, this topic proposed in view of the large measurement errors due to geometric distortion of lens imaging, and lines of automobile engine valve automatic image geometry size of image detection system and high precision of real-time online measurement requirements. Put forward the research of lens imaging distortion correction based on compress sensing, tried to use a limited step ladder shaft as the standard part, according to the camera lens imaging geometric distortion model, use limited axial standard parts to get geometric distortion discrete sampling points error value, to find a distortion correction model which can be able to achieve high precision detection for more size in a certain size range, and to be able to achieve the high-precision axial parts real-time online measurement. In this paper, the specific research contents are as follows: 1. Research the cause of image distortion, as well as commonly used several kinds of distortion correction method.2. Research Compress sensing, high-precision reconstruction algorithm and sparse signal.3. Combined with compress sensing and the geometric distortion of camera imaging model, build distortion correction model.4. Research to select suitable sparse dictionary (orthogonal basis or transform), the most sparse representation for deviation signal distortion.5. Design suitable observation matrix.6. According to the selection of coefficient of dictionary and the observation matrix, build the original deviation signal and the relationship between the distortion coefficient, and then optimization to solve the distortion coefficient. Distortion coefficient is then used to realize the geometric distortion correction of camera imaging.