Layout optimization of inspection points is key for the inspection of geometric errors in models with complex surfaces. To deal with difficulties in the adaptive planning of whole inspection points of model components that were characterized by multiple surface characteristics, mesh-based adaptive planning of inspection points was studied. A complex surface model was transformed into a dense triangular mesh model, and the triangular mesh simplification method was used for adaptive planning of inspection points. The local polynomial fitting and local curvature estimation technology were used for adaptive reduction of mesh vertexes. The vertex substitution method was used to avoid any influence from the meshing discretization error. The subset selection technology was used to improve reduction efficiency. The experimental results show that this method should not introduce the meshing discretization error of the surface, and that it was not sensitive to meshing granularity. The maximal error of the mesh-based method described here should reduce by 32.8% under the same detection points in comparison to that of the uniform sampling method, and should reduce by 16.9% in comparison to that of the random Hammersely sequence method. The mean error of the mesh-based method should reduce by 28.7% and 18.5% in comparison to that of the uniform sampling method and the random Hammersely sequence method, respectively. Meshing can avoid coordination difficulties among multiple surfaces, and the planned detection points can objectively reflect the processing quality of the complex surface.