Research On Vision-based Underwater Target Detection And Underwater Camera Calibration Method

There are abundant resources in the ocean,and exploring the ocean is of great significance.As an effective tool for exploring the ocean,underwater robots have gradually become a research hotspot.Underwater robots can replace humans in underwater operations,and even break through the limitation of human underwater operations.For safe navigation and high-quality operations,underwater robots must be equipped with accurate underwater target detection and target positioning technology.When used in a close range,sonar is susceptible to interference and its accuracy is reduced.However,the visual method is of high accuracy,but there is a problem of image quality deterioration in the underwater scene.In addition,the calibration of the camera is the prerequisite for accurate underwater positioning.Therefore,this paper studies the vision-based underwater target detection and underwater camera calibration methods for underwater robots in shallow waters.For the problem of the deterioration of underwater image quality,a dehazing method that integrates image restoration,color correction and brightness adjustment is proposed.Based on the analysis of underwater optical characteristics and imaging model,the dark channel prior is applied to the underwater scene to estimate the background light intensity and transmittance,and then the image is restored.The color restoration is based on the assumption of the gray world,and the brightness is adjusted by the gamma transformation method.After the dehazing process,the quality of underwater images gets improved.For the problem of underwater target detection,based on the characteristics of the difference between the target and the background in the underwater scene,an underwater target detection method based on visual saliency is proposed.According to the characteristics of underwater vision,the Itti model is improved and the preliminary saliency map is obtained.Considering the importance of depth features for underwater vision,the depth information is used to optimize the saliency map,and the segmentation method is used to extract underwater targets from the saliency map.For the calibration of underwater cameras,an underwater refraction camera model is established and an underwater camera calibration scheme is designed.In the process of underwater camera imaging,light has undergone multiple refractions in the water-waterproof housing-air process,so the camera model commonly used on land is no longer applicable.Based on the propagation process of light,a complete refraction camera model is established.In the refraction camera model,the parameters that need to be calibrated are the camera parameters and the housing parameters.The internal and external parameters of the camera are calibrated by Zhang's calibration method.For the calibration of the housing parameter,an error model is established and then the particle swarm optimization algorithm is used to optimize the solution.Based on the underwater robot in our laboratory,experiments of underwater image dehazing process,underwater target detection and underwater camera calibration are carried out,which verifies the feasibility and effectiveness of these methods.