Defect inspection is one of the most important tasks in industry manufacture. However, only flat products currently can be inspected by robotic camera system. My project wants to address this problem in terms of free-form shiny objects. In order to deal with this kind of shiny objects, we adopt a robotic system, consisting of a high-resolution line scan camera and a co-axis lighting. In order to meet the condition of the system, we need to divide the free-form object into different flat patches. At first, we will sample points from the CAD file and filter out some useless points. Next, we will adopt K-means, an unsupervised learning algorithm, to finish segmentation task. And then we can plan the path in different regions. In addition, we use a RGB-D camera with a novel registration method to localize the part in robot’s frame. Experiment results show that all the defects on the shiny part can be captured. With the gray scale images, we can use canny algorithm in Opencv to detect edges and then reconstruct the defects to original CAD models.