Fig. 11
From: Large scale analysis of dataset and simulation biases in SLAM research
Why do we need to include motion blur in visual tracking analysis? Motion blur is unavoidable in carefree motion with standard frame rate (30 FPS). Motion blur degrades tracking performance and in some cases causes tracking failures (two examples in the first row). Triangle indicates the starting points and circle indicates the point of track loss. Removing blur using deep de-blur71 avoids tracking failures (two examples in the first row). However, de-blurring does not necessarily improve tracking performance. In some cases (two examples in bottom row), the performance of tracking further degrades after de-blurring. The highlighted regions in bottom row demonstrate that estimated (blue) trajectory deviates from ground truth (magenta) trajectory after de-blurring while it was well aligned in the blurred version.
