Research On A Mosaic Method For UAV Images Based On Filtering

In modern society,remote sensing technology has become an effective high-tech tools for human to obtain the change of geographic information.UAV remote sensing system can overcome the limits of aerial remote sensing on long flight,bad weather,dangerous environment and so on.It can also make up the vacancy of remote sensing information of satellite area due to weather and time and provide multi-angle,high resolution images.It becomes an effective means to supplement the Aeronautical remote sensing,satellite remote sensing and remote sensing of the ground,and to avoid the problems of small work area,narrow field of view and large workload.UAV has low flight height and then can get high resolution images,but restricted by the camera focal length and flight height,the acquisition of a single UAV image can not cover the entire area of interest,in order to expand the field of view,and get more information of the overall survey area timely and accurately,we need to collect a number of smaller field of view images,and stitching them into a wide view angle of high-resolution images.As the UAV is lighter,smaller and vulnerable to high-altitude winds,resulting in flight attitude instability,and the image rotation is too large,overlapping irregular,uneven exposure,so that the stitching image prone to ghosting,existing stitching seam and other issues.In connection with these problems,this paper takes the key technology of unmanned aerial vehicle image as the research content,and tries to find a stitching algorithm which can overcome the ghosting and eliminate the stitching seam.The work of this paper mainly includes the following parts:(1)On the basis of summarizing the current research situation at home and abroad,it is found that the existing algorithms have not solved the problem of ghosting and visible stitching seam at the same time,so we propose a UAV image stitching algorithm based on filtering.The algorithm combines the weighted average fusion algorithm and dynamic programming searching optimal stitching seam algorithm effectively,it gives full play to the advantages of the two algorithms,and can avoid visible stitching seam and ghosting effect at the same time.Firstly,the UAV images are decomposed into high-frequency components and low-frequency components by Gaussian low-pass filter.Secondly,various mosaic schemes are designed to accomplish stitching process accordingly.Finally,the mosaic result is produced by linearly composing all stitching results from different components.(2)We make some improves on the weighted blending algorithm and dynamic programming search stitching line algorithm proposed by Duplaquet in the fusion rule.The improvements of the hat function weighted blending algorithm highlight the effects of the displacement in the horizontal(vertical)direction on the overall stitching results.The searching optimal stitching seam algorithm proposed by Duplaquet is improved on three aspects including the energy function,the search direction and the selection criterion of the stitching line,so that the optimal stitching line can move along the road and the edge of the house without passing through the ground.Experimental results demonstrate the effectiveness of the proposed method in avoiding visible stitching seam and ghosting effect,especially in the case of intensity difference.(3)The dynamic reference image mosaic strategy is adopted for single-band sequence images,and according to the idea of minimizing the variance of mean square variance between the same name points,the global optimization strategy is used to iterate and solve the transformation matrix of each image to the reference image,in order to reduce the accumulation of errors and achieve global optimization goals.