Target tracking algorithm based on convolutional neural network and particle filtering

Abstract

Aiming at the phenomenon that the target particle tracking algorithm is affected by illumination changes and occlusion in the target tracking process, the goal is lost. A target tracking algorithm based on convolutional neural network and particle filtering is proposed. The algorithm uses the convolutional neural network to automatically learn the depth features of the target, and extracts the more abstract semantic information of the target. The semantic information makes the algorithm robust to the apparent changes of the target, which can alleviate the drift problem to some extent. The algorithm can effectively combine the target apparent model based on convolutional neural network with the particle filter framework. The experimental results show that compared with the other five algorithms in the particle filter framework, the algorithm can track the moving targets with partial occlusion and morphological changes more robustly.