Research On Forward Looking Sonar Image Registration And Mosaic

In the process of underwater detection of FLS(forward looking sonar),its field of view is limited to a certain extent.For some large-scale scenes and targets,it is difficult to display a complete display through one detection.In this case,a single frame image cannot reflect all the information,which is not convenient for observation,and is not conducive to subsequent image processing.Image stitching technology can integrate multiple frames of images from different sources with overlapping areas into a high-resolution,comprehensive image with a global field of view.Stitching FLS images can break through the field of view limitation of FLS,and obtain FLS images with large field of view and high resolution,which can better reflect the underwater situation.This paper combines image stitching technology with the characteristics of FLS sequence images.Through the study of the key steps of image stitching technology,we design an algorithm suitable for image stitching of FLS,and integrate it into the image stitching process of FLS.The main work is as follows:The principle of FLS imaging is analyzed,and a simplified FLS imaging model is proposed,which provides a theoretical basis for image stitching of adjacent frames of FLS.The FLS maps the information of the three-dimensional space to the two-dimensional plane when imaging,and lacks the height angle information.After demonstration,this paper rationally simplified the FLS imaging model,and proposed a rigid transformation model for the splicing of images of adjacent frames of the FLS,which provides a theoretical basis for splicing.An adaptive three-dimensional block matched filtering(BM3D)denoising algorithm for FLS images is proposed.The FLS image contains a lot of speckle noise,and the existing optical denoising algorithms mostly focus on removing Gaussian noise,which does not meet the conditions of use.In this paper,power domain transformation and logarithmic transformation are performed on the FLS image,and the speckle noise of the FLS is converted into Gaussian noise with normal distribution,and then the BM3 D algorithm is used for denoising.Since the noise level of the image is unknown,this paper uses wavelet method to estimate the global noise first,and then compares it with the noise calculated by the local block to achieve the adaptation of algorithm parameters and improve the effect of uneven grayscale of the FLS image on denoising.This algorithm has been verified and is superior to median filtering and wavelet transform hard threshold filtering,and can achieve good FLS image denoising effects.An image registration algorithm based on the phase correlation method for the adjacent frames of the FLS is adopted.The FLS grayscale is uneven,the signal-to-noise ratio is high,and the features are not obvious.Therefore,the feature-based image registration algorithm is not suitable for all FLS images.For the FLS sequence image,the image overlap area of adjacent frames is large.We propose a phase-correlation registration algorithm for the adjacent frames of the FLS,using the original data of the FLS,and combining the mapping function to filter out the suitable matching For the accurate area,mask processing is used to weaken the influence of side lobes,and the amount of rotation and translation between adjacent frames is obtained,which is used for subsequent registration.The splicing process of FLS images is proposed,and the splicing algorithm is perfected according to the situation that may occur in the sonar collection process and the movement posture of the sonar.Firstly,the outliers that may appear in the sonar splicing process are filtered out,and then according to the sonar motion posture optimization algorithm,the splicing interval of the image is improved,and the image quality after the splicing is improved.Finally,according to the optimization theory of the pose map,combined with the pose of the FLS image,the straight line feature of the image at the closed loop is extracted through Hough transform,the closed loop limit is established,the cumulative error in the sonar stitching process is eliminated,and the sonar image is improved.The situation where the position drifts over time.