The processing of modern industrial parts has realized the technology for directly inputting 3D computer-aided design models into numerical control systems without the need for two-dimensional marking blueprints. The advancement of processing technology has introduced new challenges to workpiece measurement and quality evaluation methods. The profile modeling quality evaluation method based on obtaining high-precision and dense 3D point cloud models of workpieces has received attention, particularly for the manufacturing of parts with complex structures. However, the high-precision 3D measurement of parts with complex structures presents two challenges. The first is the very high surface finish of metal parts after processing. Actively projected patterns cause strong reflections on the metal surface, resulting in the problem of blindness or interreflections between complex structures, the latter of which results in aliasing. The second is the subsurface scattering of projected light owing to the translucency of composite materials. These factors induce measurement failure. The existing structured light-active vision 3D reconstruction technology cannot solve the aforementioned problems. Thus, a complex light separation model based on a single-pixel imaging method is established to address the problem of 3D reconstruction under complex illumination without spraying. This method realizes the separation of direct and complex illumination and resolves the aforementioned difficulties. To tackle the problems faced by the single-pixel imaging method, including low efficiency, speed, and its practicability in real-world measurement scenarios, a parallel single-pixel imaging method based on a local region extension method is proposed. Experiments demonstrate that the parallel single-pixel imaging method can solve the problem of 3D reconstruction under the influence of interreflections and subsurface scattering that may occur in practical measurement situations.