Target tracking strategy that combines metric learning with particle filter

Abstract

A target tracking system based on particle filtering and metric learning was proposed, focusing on the problem of the considerable decrease of target tracking performance induced by unfavorable circumstances in complex environments.To properly obtain the target features, a convolutional neural network (CNN) was first offline-trained using the suggested method.Then, using the metric learning for kernel regression (MLKR) approach, the distance measurement matrix optimization model to minimize the prediction error could be built. The resulting model could then be handled by applying the gradient descent approach to obtain the best solution of the candidate target.Additionally, the reconstruction error was computed to build the target observation model based on the anticipated value of the best candidate target.