This seminar will consist of two parts.
In the first half, I will present a vision-aided navigation system currently in development at CMU. We combine a stereo visual odometry system with an aided inertial navigation filter to produce a precise and robust system that does not rely on external infrastructure. I will present accurate results from data acquired in rural and urban scenes on a tractor and a passenger car traveling distances of several kilometers.
In the second half, I will present a new approach for vanishing point and
Manhattan directions estimation. Being a typical problem of multiple model estimation I will discuss the most common algorithms for this task and how they have been applied to vanishing point estimation. Furthermore, I will present an algorithm called J-Linkage recently introduced by Toldo and Fusiello. I will show that it is a very good choice for implementing a simple and efficient vanishing point detection algorithm. The most important strength of this approach is that it is able to rely on an error metric in the image rather than on the so-called Gaussian Sphere. The source code of the whole vanishing point detection pipeline is available online.