Improved CACF Algorithm Based on Channel Reliability

Correlation filter (CF) based trackers has become the mainstream tracking algorithm due to their well performance both on accuracy, precision and tracking efficiency, recently. Context-aware correlation filter (CACF) reformulates CF tracker by introducing background supervision in energy function, which gets excellent performance in handling background clutter. However, CACF is not robust when the target is occluded. To remedy this, a pre-training network is used to extract the depth features, and then the weight factor is computed according to the contribution of different channels. Then, the model is updated adaptivly based on the output response map. In summary, an improved CACF algorithm is proposed by combining channel reliability (CR-CACF) and adaptive model updating. Both qualitative and quantitive experiments on public benchmark prove that the proposed method get excellent performance compared with some state-of-the-art tracking, and reduce dimension effectively reduce computational complexity.