3D Scene Description Based On 2D Sonar Image

Imaging system plays a key role in underwater operation.Computer processing of forward-looking sonar video image has been widely used in turbid underwater environment.In various applications,using acoustic image to understand scene features and identify underwater objects has become a very important problem.Sonar image is formed by emitting acoustic pulses and measuring the acoustic power reflected from the scene surface.Interpreting the three-dimensional world based on the visual clues in the two-dimensional forward-looking sonar image is the recovery process of the lost elevation information in the geometric projection process of sonar imaging.On the one hand,the research content of this paper is to model the imaging process of high-resolution two-dimensional forward-looking sonar system and recover the unknown pitch angle from a single frame sonar image.On the other hand,the 3D object is reconstructed according to the matching feature points in the sonar image from multiple sonar angles.In this paper,the image formation process of forward-looking sonar is modeled first,and then a method to recover the unknown pitch angle according to the target shadow and target brightness in the image,and then reconstruct the three-dimensional object is studied.This method is suitable for single frame forward-looking sonar image.It is assumed that the target has Lambert surface,and its surface changes monotonically with the distance from the sonar,And cast visible shadows on a flat background.Then,the experimental results with real pool data are given to verify the performance of the three-dimensional reconstruction technology based on target contour and gray scale.The reconstruction steps include image preprocessing,image feature extraction and target area reconstruction.In the preprocessing part,the average amplitude gain compensation method and median filter are used to enhance the image recognition.In the part of feature extraction,threshold segmentation and morphological methods are used to extract the feature region,and then the edge is calibrated.For three-dimensional reconstruction under multiple sonar angles,the epipolar line is an important basis for matching image feature points.This paper discusses the epipolar geometric constraints under multiple sonar angles,analyzes the different angles produced by different sonar movements and the influence of noise on the epipolar line,and simulates and analyzes the reconstruction effects of two mathematical analysis methods,Two reconstruction methods of epipolar constraint search synchronization are studied and simulated.The experimental results of simulated data are given to verify the effectiveness of the epipolar constrained search synchronous mean reconstruction method.