ردیابی بصری قوی از طریق ویژگی های چند لایه
Abstract
I. Introduction
II. An Improved Convolutional Neural Network Based on RPReLU and the SVM-Dropout Method
III. Correlation Filter Based on SGA
IV. RSCNN-Based SGACF
V. Experimental Analysis
Authors
Figures
References
Abstract
For problems related to the robust tracking of visual objects in various challenging tracking conditions, a robust visual tracking method based on multilayer convolutional features and correlation filtering is proposed. To solve the problems of mean deviation and insufficient discrimination ability in traditional convolutional neural networks (CNN), this study proposes randomized parametric rectified linear units (RPReLU) as the activation function. Meanwhile, the zero-setting operation of weights in the traditional dropout process occurs randomly and fails to discriminate the features with different weights, which leads to a low learning efficiency. Therefore, this study proposes an improved dropout method based on a support vector machine (SVM), which provides a selective dropout rate to increase the manual orientation and improve the learning efficiency of the dropout process. In addition, traditional CNN trackers only employ the output of the last layer, which can effectively capture semantic features but not spatial features. To solve this problem, we propose to use the rich features of the multiple convolution layers of CaffeNet as the target representation. Furthermore, we propose an improved correlation filter to further improve the tracking performance and improve the tracker’s capability of dealing with scale changes, which effectively solves the problem of adaptive estimating of target size. The extensive experimental evaluations have been carried out through the OTB2015, VOT2016 and VOT2018 datasets, proving that the proposed method is very effective in dealing with a variety of challenging factors.
Introduction
Visual target tracking is a valuable research that has been widely used in frontier fields such as traffic accident supervision, automatic driving, intelligent home, and weapon control [1]–[4]. While much effort has been expended to develop the robustness and efficiency of visual trackers, target tracking still needs to address the following major challenges: 1) various interference factors, including low resolution, rotation, scale change, occlusion, deformation, motion blur and so on; 2) insufficient tracking efficiency, accuracy and stability [5]–[7]. Therefore, the main task of this research is to solve these two problems. In recent years, a large number of visual tracking methods have been proposed to solve target tracking problems. An et al. [8] proposed a mean shift tracking algorithm based on 3D colour histogram. This method deals with the influence of a low-lighting environment and similar targets on tracking. However, when the intensity or distribution of light changes dramatically, its tracking effect is not ideal. Zhou et al. [9] proposed a tracking method that can not only suppress background interference, but also increase foreground weight by using foreground probability and candidate model weight histogram. This method solves the interference of background change and illumination change, but the tracking failure rate is high under the interference of target occlusion and motion blur.