Simulator-aided Edge-based Acoustic Camera Pose Estimation

Abstract

In this work, a method to estimate accurate 6 degrees of freedom acoustic camera pose is proposed. Acoustic cameras, also known as 2D forward-looking sonar can generate high resolution 2D images in all water bodies. However, due to the unique imaging principle, techniques such as vision-based localization with acoustic images are still at the early stage. Even we have full 3D information of the scene, it is still difficult to localize the sensor. In this work, we deal with the problem of estimating the accurate relative pose between the camera and a 3D known target which can be applied to tasks like extrinsic calibration. Previous works mainly estimate the camera pose with planar targets; however, the methods are not robust to noise in the image. We propose an accurate pose estimation method with the help of an acoustic image simulator which can deal with 3D targets. We use edge features to match the real images and the synthetic images. A coarse-to-fine strategy is used for global localization and pose refinement. Experiments prove the effectiveness of the method.