Intelligent Track Association Method Based on Deep and Composite Features

Track-to-track association plays an important role in multi-sensor tracking and fusion system. The correct rate of association will directly influence the accuracy of tracking in multi-sensor system. However, when the number of targets increases, using only traditional motion features could lead to a decrease in the correct rate of association due to their insufficient representation capability. In this paper, the fuzzy based track association is improved by using deep features of tracks extracted by neural network and motion model probability obtained with Interactive Multi-Model (IMM). A Convolutional Neural Network (CNN) is constructed with residual structure to extract deep features of tracks, and the fuzzy association factors is redesigned by deep features and model probability in this paper. The effectiveness of the proposed method is illustrated by various simulation scenarios. Experimental results demonstrate that the proposed method has better performances compared with traditional methods.