For robotic assembly long cylindrical shaft and countersunk-hole parts， this paper investigates the assembly strategies of combining 3D vision， and force perception for compliantly assembling parts in arbitrary positions and poses during assembly. An algorithm is proposed based 3D vision for pose estimation of countersunk-hole parts： firstly， use supervoxel segmentation， Constrained Planar Cuts （CPC）and clustering to complete the preprocessing of point cloud， and then use Weight-based Random Sample Consensus （WRANSAC） plane fitting to acquire plane pose， finally， estimate the assembly position by profile extraction and circle fitting of countersunk-hole. In addition， we implement the compliant assembly of the parts using impedance control. The experimental results indicate that the average distance error ∆d of WRANSAC plane fitting algorithm can be as high as 0.09 mm and the angular error ∆θ can be as high as 1.8°. Moreover， the angular error of the countersunk-hole position in 3D space is 1.0°， which can realize the assembly of parts with a mating clearance of 0.5 mm. The assembly strategy meets the requirements of high precision and compliance of countersunk-hole assembly.