Search for a command to run...
In this paper we describe a method that estimates the motion of a calibrated camera (settled on an experimental vehicle) and the tridimensional geometry of the environment. The only data used is a video input. In fact, interest points are tracked and matched between frames at video rate. Robust estimates of the camera motion are computed in real-time, key-frames are selected and permit the features 3D reconstruction. The algorithm is particularly appropriate to the reconstruction of long images sequences thanks to the introduction of a fast and local bundle adjustment method that ensures both good accuracy and consistency of the estimated camera poses along the sequence. It also largely reduces computational complexity compared to a global bundle adjustment. Experiments on real data were carried out to evaluate speed and robustness of the method for a sequence of about one kilometer long. Results are also compared to the ground truth measured with a differential GPS.