Author(s)

Xiaoyu Guo, Haoyu Liu and Hengzhang Dai

Corresponding Author

Xiaoyu Guo

ABSTRACT

In order to solve the problem that intelligent mobile robots often lose tracking targets due to changes in the appearance of targets in the tracking process, Caffe deep learning framework and ROS robot operating system are used as the development platform. A mobile robot target tracking system with high accuracy and high real-time performance is designed and studied. Using GOTURN target tracking algorithm based on twin convolution neural network, which is robust to target deformation, viewing angle, slight occlusion and illumination change, the tracking model of offline training is applied to TurtleBot mobile robot in real time through ROS system, and detailed tests are carried out. The target tracking system not only has a feasible design scheme and can effectively track targets in various complex scenes, but also has the characteristics of low cost, high performance and easy expansion.Abstract: In order to solve the problem that the current intelligent mobile robot often loses the tracking target due to the appearance change of the target during the tracking process, we have designed a high accuracy and high Real-time mobile robot target tracking system. We use a GOTURN target tracking algorithm based on twin convolutional neural networks that is robust to target deformation, viewing angle, slight occlusion, and illumination changes. We used the ROS system to apply the offline training tracking model to the TurtleBot mobile robot in real time, and conducted detailed tests. The experimental results show that the target tracking system not only has a feasible design scheme, but also realizes that the mobile robot can effectively track the target in various complex scenarios. In addition, it also has the characteristics of low cost, high performance and easy expansion.

KEYWORDS

Target tracking; Caffe deep learning framework; twin convolution neural network; GOTURN target tracking algorithm; ROS