Vision-based motion estimation for vehicles on test track via cone markers

In this paper, a vision-based motion estimation algorithm based on the ORB feature detector is presented. The mono camera is mounted facing forwards and the method is not utilized any aprior motion model. Supplementing the general detection-match-pose estimation steps, an iterative loop is proposed to eliminate the problems of the vision-based technique and increase the accuracy. The method operates with vehicle dynamic based assumptions since the sensor is mounted to a car-like ground vehicle. The motion estimation results can be used for orientation smoothing, sideslip estimation, or sensor calibration.