Abstract. The problem of recognizing 2D objects from 3D models is a well known problem in computer vision. A lot of methods have been developed during the last years. However, most of these methods work on natural images, and few results are available for stroke images, as for instance storyboards. This research work is motivated by the desire to automatically reconstruct the 3D scenes from the 2D storyboards of a company producing 3D animated films. In the case of this project, the storyboards are hand drawn by the artists in charge of waiting the scenario of the movie. Then, modeling specialists create the 3D models of the different scenes based on the storyboards using 3D object models stored in a database. Our research aims to recognize each 3D model from the corresponding piece of sketch (in the storyboard plan), along with its 3D viewing angle, its scale and its rotation angle in the drawing plane, so that it can be automatically and correctly placed in the 3D scene. An additional problem that arising here is to recover the camera position and the scene most similar to the input storyboard. The planned results will help the company economize valuable time spent on reconstructing the 3D scene.