Improved Hierarchical Convolutional Features for Robust Visual Object Tracking

Figure 1

Algorithm model. In frame t, the blue box presents the prediction result of the previous frame and the green box presents the searching area with 1.5 padding. The convolution features are extracted from the original image within the green box. In frame t + 1, features for the prediction are extracted around the location in the previous image. The response of the candidate region is calculated by the trained filter , and the position with the maximum response value is the predicted target position.