Processing and Forecasting the Trajectory of a Thrown Object Measured by the Stereo Vision System

Transportation of small-sized rigid objects in industrial environment may be provided by throwing it from the source point and catching it at the destination point. This approach promises better flexibility than traditional transportation systems based on conveyor belts. Accurate real-time forecasting of the object ballistic trajectory is necessary to provide successful catching of the object by the gripper. The development of a sample-based algorithm for trajectory forecasting is a scope of this paper. The input for the forecast is a reference of object spatial coordinates measured by the stereo vision system. Such measurements allow defining the position of the object in a camera-related coordinate system with millimeter accuracy, however they sometimes include outliers. A reference of coordinate transformations is proposed, which translates object coordinates from the camera related 3D system to a 2D system with relations to gravity and motion direction. Outlier detection is made during these transformations. The forecasting is performed in 2D coordinate system with use of k nearest neighbors approach. Applying the algorithm to the measured trajectories showed that it is able to predict future position of the object with 3 centimeters precision in 92 % of cases.