Research On Multi-view Motion Deblurring Method Of Moving Target

The three-dimensional vision inspection is an important technology to achieveautomatic and intelligence in the field of manufacturing industry. But in thepractical production line, a motion blur image will be obtained by camera becauseproduct is in motion state. It is difficult to use these motion blur images forthree-dimensional reconstruction and it is not possible to realize the visual inspection.Therefore, these motion blur images will be needed to restore. However, the existingmethods of restoration have their limitation. For example, the corresponding relationof each image feature point will destroy while the blur image is restored by themethod of single image deblurring, and the multiple images deblurring methodrequires the background of blur images nearly the same as others. These limitationsmake these restored images are difficult to use for three-dimensional reconstruction,then they also could not realize the three-dimensional vision inspection. Tosolve this problem, a new image deblurring method based on multi-view images ofmotive target is proposed.In this paper, the proposed image deblurring method not only improvesthe clarity of motion-blurred images from different angles, but also ensures thecorresponding relation of point spread function of the motion-blurredimages from different angles. In order to ensure the corresponding relation of thesame characteristics from each image, we have to establish the relationship of the pathof multi-view image point spread function. First of all, the point spread function ofsome motion-blurred images, taking two motion-blurred images as an example, isestimated. Secondly, the parameter matrix of each camera is obtained by a cameracalibration method based on multi-view geometry theory. In the combination withestimated point spread function and camera parameter matrix, the relationship of pointspread function between motion–blurred images from different angles is built, takingthe example of linear motion. The correspondence to multi-view blurring path isguaranteed based on least square error rule, and the point spread function of eachmotion blurred image is optimized accurately. At last, multi-view motionblurred images are restored as a whole.A series of experiments show the effectiveness of the proposed method based on multi-view images of motive target. In the process of the experiment, the main stepsof this method are verified by experiments in the first place. Then some restorationtests are conducted for the simulating blur images in the situation of differentathletic forms. At last, depending on the different numbers of blur image, somerestoration tests are conducted for the real blur images in the situation of differentathletic forms. The experimental results show that the proposed method could realizemultiple images deblurring which have different background. Meanwhile, this methodcan also restore the real blur images caused by different forms of motion.