Multi-scale Correlation Filtering Visual Tracking via CNN Features

In order to meet the requirements of tracking accuracy and speed of visual tracking algorithm, a multi-scale correlation filtering visual tracking algorithm combined with CNN features is proposed. For the challenge of the training for convolutional neural network consumes numerous data and time we exploit an online convolutional neural network training method that extracts the features of the target context through a shallow network layer. Then, the correlation filtering algorithm is applied to the visual tracking of the given target features and the multi-scale search of the optimal response. Finally, the experiment results show that only using two simple network layers to extract the target features as a multi-channel feature of kernel correlation filtering can achieve excellent results.