Convolutional regression for visual tracking

by chkap 2017-03-23

In this blog, I introduce one of my papers on visual object tracking.

The initial version of this paper is submitted to arXiv.org.

Abstract:

Recently, discriminatively learned correlation filters (DCF) has drawn much attention in visual object tracking community. The success of DCF is potentially attributed to the fact that a large amount of samples are utilized to train the ridge regression model and predict the location of object. To solve the regression problem in an efficient way, these samples are all generated by circularly shifting from a search patch. However, these synthetic samples also induce some negative effects which weaken the robustness of DCF based trackers.
In this paper, we propose a Convolutional Regression framework for visual tracking (CRT). Instead of learning the linear regression model in a closed form, we try to solve the regression problem by optimizing a one-channel-output convolution layer with Gradient Descent (GD). In particular, the receptive field size of the convolution layer is set to the size of object. Contrary to DCF, it is possible to incorporate all "real" samples clipped from the whole image. A critical issue of the GD approach is that most of the convolutional samples are negative and the contribution of positive samples will be suppressed. To address this problem, we propose a novel Automatic Hard Negative Mining method to eliminate easy negatives and enhance positives. Extensive experiments are conducted on a widely-used benchmark with 100 sequences. The results show that the proposed algorithm achieves outstanding performance and outperforms almost all the existing DCF based algorithms. 

 

The contributions in this paper are summarised as follows.

  1. This is the first paper that proposes to solve the regression problem for visual tracking via single convolution layer. The new approach eliminates the inherent defect induced by the circular structure in the famous DCF approach.
  2. We propose a novel robust loss function to speed up the training and detecting stages. It also significantly improves the performance as shown in the experiments.
  3. We perform extensive experiments on popular benchmarks. And the results show that, our proposed algorithms outperforms most of the existing DCF approach, including the famus SRDCF and its variants like DeepSRDCF and SRDCFdecon.

 

The overall results are shown in the following figure.

The results on OTB-100

 

Here is a demo video of the proposed tracker.