Ball Trajectory Prediction Using a Camera-Inertial Sensor
This page contains the results of the conducted experiments to see how well the flight of balls on a soccer field can be predicted using a camera-inertial sensor. The results are given as a set of videos visualizing the prediction performance of time and were initially presented in . Please note, that we wanted to prove that these sensors in principle provide enough information for such a prediction. So this study uses manually detected circles as an input to the estimation and prediction layer.
The videos are splitted into four quadrants:
- On the top-left a three-dimensional model showing the objects of interest recreated from the estimated states. The camera and the ball, as well as the predicted trajectory will be visible. Additionaly, the actual bouncing point is shown as a red circle on the ground.
- On the top-right right the image as seen from the camera is reconstructed from the estimated state.
- On the lower-left the actual camera image is shown. It is augmented by showing the predicted ball trajectory as purple circles and an artificial horizon.
- On the lower-right a graph is shown which depicts the prediction error over time.
Ball Moving Towards the Observer
Ball Passing by the Observer
Long-Distance Ball Flight Towards the Observer
Spectacular goal from television perspective
The introduced methods can be used to estimate and predict the state of ball and camera using only the images from a television broadcasted scene. The following video shows how this might look like. Because of the poor quality of the television images and the partially unknown dimensions of the soccer field, the perfomance is not as good as in the scenes above, but indicates, that the method is well suited for this kind of problem, too. Download high-quality version
Experimental Data and Source Code
The experimental data and the source code leading to this results will be made available for download in the near future.