Study On Technique Of Displacement Measurement Based On Motion Blurred Image

With the development of science and technology, the requirement of motion measurement is higher and higher, in terms of measurement accuracy, efficiency and cost. The machine vision technology got eager attention because of its non-contact measurement, simple equipment, low cost. This thesis researches the methods of estimating displacement from the motion blurred image by means of frequency domain, the alpha channel and large displacement optical flow.Firstly, proposing a method of displacement estimation based on the frequency domain features of a single color motion blurred image. The Fourier transform of the motion blurred image is analyzed, thus the motion information for motion estimation in blurred image can be obtained. We treat the three channels separately for color image’s motion estimation to avoid the loss of information by image rotation. In order to avoid the error of parameters getting stacked in motion estimation, estimating the length and orientation of the motion respectively. The orientation can be estimated from the response of the Gabor filters which convolve with the spectrum of the blurred image. Additionally, the motion extent can be obtained using the cepstrum method. The experimental results show that this method is simple and effective. It is reliable for the motion parameters estimation based on one single blurred image.Secondly, for the problem of complex calculation and large consumption of motion estimation from blurred image, a novel method of estimating the motion parameters based on alpha channel of corner region is proposed. The motion blurred image contains motion information, and the corner motion is a significant interpretation for motion parameters estimation. The alpha channel motion blur constraint of the corner region is established, which can separate the motion estimation and image deblurring; then taking the Hough voting approach to extract the information which we need, to get the motion estimation. The gradient calculation of alpha channel is directly related to the filter, analyzed the effects of different filters on estimation accuracy. The experiment results show that the method is effective within a certain motion range by using a single blurred image, the operation time is shortened and estimation result is accuracy.Eventually, in view of the traditional optical flow algorithm is dependent on clear image sequences, and the effect on large displacement estimation is poor, put forward using the information from the motion blurred image to improve the optical flow estimation precision. By using alternating exposure model, introducing a motion-blurred-image between two clear images to estimate the optical flow field. In order to restrain the influence of light changes, take advantage of texture-part of the image by using the structure-texture decomposition approach. Due to the L1 norm is more advantage than the L2 norm in continuous and piecewise smooth, so the TV-L1 optical flow model based on the motion blurred image is established. Take the simulation experiment in Matlab platform, we get the experimental results and analysis; confirm the method is benefit to improve the accuracy of the large displacement optical flow.