The article presents our solution for the classification of moving flying objects in a video sequence captured by a static camera. The tool uses the extraction of scale and rotation invariant SIFT features, which allow the multi-class SVM to classify the examined object into one of the considered classes: 'bird', 'plane' or 'negative'. The most successful of our tested models achieved accuracy of over 90% and their recall and precision for each class reached values above 90%.