Integration of regularized l1 tracking and instance segmentation for video object tracking

We introduce a tracking-by-detection method that integrates deep object detector with particle filter tracker under the regularization framework where tracked is represented by sparse dictionary. A novel observation model which establishes consensus between and formulated enables us to update dictionary guidance of detector. This yields an efficient representation appearance through video sequence hence improves robustness occlusion pose changes. Moreover we propose new state vector consisting translation, rotation, scaling shearing parameters allows tracking deformed bounding boxes significantly increases scale Numerical results reported on challenging VOT2016 VOT2018 benchmarking data sets demonstrate introduced tracker, L1DPF-M, achieves comparable both while it outperforms state-of-the-art trackers improvement achieved in success rate at IoU-th=0.5 11% 9%, respectively.