Abstract: A target recognition and registration algorithm based on domain knowledge in 3-dimensional dynamic scene is proposed to enhance the sensory immersion of the teleoperation system and make the operator better integrate into the remote working environment. Firstly, a domain knowledge database containing multi-view point cloud features and assembly constraints is constructed by offline parsing and segmentation of virtual prototype CAD (computer aided design) models. Then, the CVFH (clustered viewpoint feature histogram) and FPFH (fast point feature histogram) features are computed by dynamically collecting scene point clouds, and the multi-view point cloud features contained in domain knowledge database and the CVFH features are compared to implement target recognition. The target pose is obtained using the FPFH features through two-step registration. Finally, the accurate registration and real-time push of guidance information driven by the changes of working state of the teleoperation robot are implemented using the assembly constraint knowledge database. The experimental results show that the algorithm can not only effectively guide the remote robot to complete the maintenance operation, but also improve the precision and efficiency of teleoperation.