Error reduction for large rotational motion estimation of autonomous vehicle

This paper presents a methodology for reducing error in large rotation using an omnidirectional camera based on bundle adjustment technique. One of the most traditional problems of motion estimation method is accumulative error. The advantage of omnidirectional camera is that allows tracking landmarks in large rotation and long travel. The minimal error is applied on set of sequent images so that landmarks are tracked in all images. The global motion is estimated in high accuracy based on utility to optimize the partial rotation error based on the bundle adjustment technique. The criterion for evaluation is the angular deviation between projection and back-projection of corresponding points. This criterion makes advantage to use quasiconvex optimization method for solving problem. The experimental equipment consists of an electric vehicle with an omnidirectional camera, laser sensor and GPS receiver. The data was collected under large rotation and different terrain in outdoor environments. In order to evaluate proposed method, the vehicle positions were compared with GPS information and plotted on satellite images from Google Maps.