Accuracy of visual ego-motion estimation: A bike mapping system case study

This paper describes first tests of mobile mapping system prototype designed for a bike. The mapping platform is equipped with two cameras synchronized using the GNSS receiver output. Other devices such as a laser scanner or IMU are supposed to be fully integrated in the future. Addressed research concentrates on the accuracy of a trajectory estimation using purely visual information. During the tests, three sequences of over 200 images were acquired along with GNSS real-time measurement. Afterwards the trajectory parameters were estimated both in the closed-loop and opened-loop scenarios. The reference trajectory was determined using bundle adjustment with control points. Additionally the coordinates of the trajectories estimated using purely visual approaches were compared with those resulting from GNSS measurements. Results show that considerable drift of parameter estimates observed for purely visual method can be reduced in the large extent providing the proper loop closure.