Research And Application Of Rigid Body Pose Estimation For Water Conservancy Dam

6Do F pose estimation of dam structure has broad application prospects in the field of water conservancy,and is a fundamental requirement for dam surface defect localization and augmented reality applications.Some scholars have realized the augmented reality application of dam body construction monitoring and filling process visualization,but the dam pose estimation in the process of 3D registration is realized by manual calibration.6Do F pose estimation technology based on a single RGB image calculates the 6Do F pose of the camera relative to the target object in real time from the image captured by the mobile device,which enables the camera to be located quickly and accurately in the real world.At present,6Do F pose estimation for desktop objects is a hot spot in application and research,but its technical route and dataset production method are not suitable for water conservancy dam application scenarios.Meanwhile,6Do F pose estimation technology for outdoor applications is mainly based on high-precision IMU and GPS hardwares or rely on 3D point cloud map to complete camera pose estimation,which has a high application cost.Aiming at the features of dam applications including inconvenient camera pose truth marking,serious perspective deformation of dam target,on-site vegetation sheltering,complex environment and unevenly designated keypoints,and considering the application requirements of defect locating on dam surfaces,the work of this paper is as follows:1.A 6Do F pose estimation method for dams considering features of dam sites is leveraged.Firstly,a modified HG-Stacked model is used to locate the keypoints of the dam,secondly the dam pose is solved by optimizing the weighted reprojection error of keypoints from the initial pose,and finally a method to form the 6Do F pose estimation dataset of dams for our pipeline is presented.In the experiments using the dataset of a concrete arch dam,the identification accuracy of our keypoint detector is about 1.4% higher than the one with the highest identification accuracy among the comparison methods in this paper.The pose estimation method leveraged in this paper can further improve the accuracy based on the initial pose.Our proposed dataset production pipeline suits dam applications well and the 2D coordinates of keypoints can be marked quickly from video sequences.2.Based on our dam pose estimation method,we leveraged a coordinate recognition algorithm of dam surface,and finally a dam surface defect localization system is designed and implemented with the requirement of manual inspection.From on-site pictures taken by dam inspectors,the system calculates surface coordinates of defects found while the estimated dam pose is shown in AR mode.Experiments on an arch dam shows that the average error of dam surface elevation is 0.212 meters,and the accuracy rate of dam surface pile number(accurate to integer)recognition is 96.3%,which meets the requirements of automatic dam surface coordinate recognition application.The testing of the dam surface defect localization system shows that the system operates well and can meet business requirements.The 6Do F pose estimation method proposed in this paper has good applicability and high pose estimation accuracy in dam scenarios.The dam surface defect localization system is convenient to use and helps dam inspectors quickly assess coordinates of damaged engineering parts.Therefore,the content of this paper has certain theoretical significance and application value.