Proc. International Conference on Computer Vision, 1995.
We present a probabilistic 3D object recognition algorithm which uses the probabilistic peaking effect of measured angles and ratios of lengths to guide the recognition process. Isoratio and isoangle curves are traced on the viewing sphere and are used to compute conditional probabilities. These probabilities are incorporated into a probabilistic model which deals with various types of uncertainty associated with the recognition process, such as uncertainty due to imperfect results of edge and corner detection (e.g. detecting edges which belong to the background, not detecting edges which belong to the object, and uncertainty in edge and corner position). We perform pose estimation and compute the uncertainty in pose due to the uncertainty in measurements in the image. The uncertainty pose subspaces are used to find hypotheses which reinforce each other. We compute the probability that these sets of hypotheses are correct and use these probabilities to rank the matching hypotheses.