Multi-target Tracking Algorithm Based on Convolutional Neural Network and Guided Sample

摘要

In order to reduce the number of samples when using the convolutional neural network to train the moving target template online and improve the validity of samples, a sample selection method based on guided samples is proposed and applied to the fast multi-domain convolutional neural network tracking algorithm. The basic idea of the sample selection method is as follows, the initial samples are determined firstly by the sample filtering method of frame level detection and nonlinear regression model, and then the similarity between the initial samples and the target template are calculated, the samples with the similarity greater than a certain threshold are finally used as the guidance sample. The experimental results show that the tracking time of the proposed tracking algorithm is greatly reduced compared with the fast multi-domain convolutional neural network, the proposed tracking algorithm can speed up the tracking speed, improve the accuracy and robustness in complex environments.