Inline, also known as on-line, measurement is one of the most important measurement techniques in automotive manufacturing industry, especially for body shop. Nowadays, optical measurement systems, which is most commonly used for inline measurement, are developing rapidly. However, some of dimensional features in body shop such as spot welding points, stud welding, high reflective surfaces and black painted materials, are still challenging for traditional optical inline systems to measure. Aiming at optical challenging dimensional features, a measuring system with cameras and fixed illuminations is developed to achieve robust and accurate measurement. Photometric stereo technique is used to obtain the surface normal map of the feature, so the algorithm can adapt to various reflections of the material. A Histogram of Oriented Normals (HON) extractor is proposed to extract the feature vector, which is small enough to fit a neural network, from normal map. After building a database with 1000 normal map and corresponding feature position, an artificial neural network is trained for localization the feature from 2D image. Combining 2D information from two different cameras, the 3D coordinate will be available with triangulation. Comparing with traditional HOG bench mark extractor, proposed system showed significant advantage on optical challenging materials in the experiment with welded stud sample. Comparing with off-line measurement systems, proposed system takes much less time, which gives the potential for optical challenging features inline measurement.