The Matrioska Tracking Algorithm on LTDT2014 Dataset

We present a quantitative evaluation of Matrioska, a novel framework for the detection and tracking in real-time of unknown object in a video stream, on the LTDT2014 dataset that includes six sequences for the evaluation of single-object long-term visual trackers. Matrioska follows the approach of tracking by detection: the detector localizes the target object in each frame, using multiple keypoint-based methods. To account for appearance changes, the learning module updates both the target object and background model with a growing and pruning approach.