Research On Stereo Matching Of Vibration Degarded Image And Its Application

Binocular stereo vision system working platform is not absolute still, mechanical vibration of vision system working platform exists more or less. Los vibration will cause relative motion between shooting target and camera, resluting in shooting target deviated from the original imaging location in the imaging process. Blur will occur in the image so that the binocular image can't do stereo matching.In order to simulate the vibration disturbance of binocular stereo vision system in practice, designed and built a vibration experimental platform that vibration form is controllalbe to collect vibration degraded image. Use the vibration blurred image charater in frequency domain that a set of dark stripes that appears in specrum of blurred image corresponding to zero values of the zero order Bessel function to identify the vibration blur parameters. The degradation mechanism of vibration blurred image is also analyzed.On the basis of the above, a novel line-by-line algorithm with considering the impact of adjacent images is proposed. Specific practices are as follows: take ith image to be restored as center, get N line images up and down, the value of N is equal to the vibration length. Use PSF to restore selected image, then take center line of restored image as ith image. Repeat the above process. Apply line by line algorithm with considering the impact of adjacent images to restore the vibration degraded image and use the MSE and PSNR to evaluate image quality.The results show that image restored by line-by-line algorithm with considering the impact of adjacent images has the smallest MSE and the biggest PSNR, and gain much clearer details.Use the parameters gained by calibration to correct the binocular images. The disparity map is obtained by the stereo matching, and the 3D coordinates of the target corner are calculated by using the disparity map. The results show that the effect of stereo matching is good, and the 3D coordinates of corner points in the image can be obtained acurately.