To address the problems of viewpoint redundancy in the automated visual inspection of complex workpieces and degraded viewpoint quality induced by occlusion in binocular imaging, a binocular vision inspection planning method based on surface curvature is proposed. First, the number and pose of sampling viewpoints are optimized according to the curvature distribution and the direction of maximum curvature of the mesh model of the scanned part. An improved two-step screening greedy algorithm is developed to select the optimal viewpoint set under the condition of uneven meshing, and the simulated annealing algorithm is adopted to optimize the viewpoint visiting sequence, thus generating the shortest collision-free visual inspection path for the camera. Verified by both simulation and physical experiments, compared with traditional inspection planning methods, the proposed method reduces the number of viewpoints by 37% and shortens the inspection path by 26.56% while satisfying the surface coverage requirement, which effectively improves the scanning efficiency and accuracy in automated visual inspection scenarios.