Intelligent Target Tracking of Unmanned Vehicles Considering Convolutional Neural Networks

Abstract

The target tracking system and remote control system of unmanned vehicles have been paid more and more attention by major automobile manufacturers. As mobile Internet giants such as Apple and Google begin to combine mobile devices, mobile systems and in-vehicle systems, this has revolutionized the intelligent system of unmanned vehicles. However, target tracking systems and remote control systems for unmanned vehicles are in their infancy, especially when implemented on mobile devices. Aiming at the difficulties in tracking tasks, this paper proposes a tracking algorithm based on deep learning. First, offline pre-training is performed on tens of thousands of general target images through a convolutional deep neural network. Since the types of pre-trained images are extensive and the features used are structural, when the target is occluded and changed, the changed target can still be re-represented by retraining the parameters of the feature. The tracking method based on convolutional neural network and particle filter framework provides a good experimental framework for solving problems such as target changes. The tracking method proposed in this paper shows good tracking ability in multiple test video sequences with good real-time performance and accuracy.