Joint learning hash codes and distance metric for visual tracking

In this paper, we propose a novel tracking algorithm based on joint learning hash codes and distance metric. We formulate the visual tracking as an Approximate Nearest Neighbor (ANN) searching process in which hashing methods have achieved promising performances. But most existing hashing methods rely on an affinity or similarity matrix measured by simple Euclidean distance. To obtain more robust hash codes for tracking, we utilize distance metric learning method to measure the similarity. We propose a joint learning hash codes and distance metric algorithm for visual tracking and a fast solution is developed to solve these two problems simultaneously by cross gradient descent. Then we use the learnt hash function to encode the templates and candidates and conduct the ANN searching. Extensive experiments on various challenging sequences show that the proposed algorithm performs favorably against the state-of-the-art methods.