Using Spin Images for Efficient Object Recognition in Cluttered 3D Scenes
IEEE Transactions on Pattern Analysis and Machine Intelligence
SMI 2011: Full Paper: Geometric models with weigthed topology
Computers and Graphics
Detection, classification and estimation of individual shapes in 2D and 3D point clouds
Computational Statistics & Data Analysis
| Hi-index | 0.00 |
A robust and efficient method is presented for recognizing objects in unstructured 3D point clouds acquired from photos. The method first finds the locations of target objects using single spin image matching and then retrieves the orientation and quality of the match using the iterative closest point (ICP) algorithm. In contrast to classic use of spin images as object descriptors, no vertex surface normals are needed, but a global orientation of the scene is used. This assumption allows for an efficient and robust way to detect objects in unstructured point data. In our experiments we show that our spin matching approach is capable of detecting cars in a 3D reconstruction from photos. Moreover, the application of the ICP algorithm afterwards allows us (1) to fit a query model in the scene to retrieve the car's orientation and (2) to distinguish between cars with a similar shape and a different shape using the residual error of the fit. This allows us to locate and recognize different types of cars.