The effect of motion blur on the ManakinTracker

The Manakin Tracker was developed to track a small tropical bird, called golden-collaredmanakin, which performs its courtship dance at a very high speed. The fast movement leads to strong motion blur. The ManakinTracker is based on a Convolutional Neural Network(CNN) as well as blob detection through background subtraction based on a Mixture of Gaussians model. The CNN was trained on images cropped from frames in a data set of videos depicting the golden-collared manakin displaying its courtship dance. In our experiments, we pre-processed (simulated motion blur, rotated, deblurred) the videos toassess how motion blur affects the performance of the ManakinTracker. We found that when we simulated motion blur, less motion blur lead to an increased robustness of the ManakinTracker.