Object Tracking by Filtering Neuron Network

Abstract

Object tracking simulation using neural network is described. Since control value for tracking should be determined in accordance with past state, such model is needed that can process time series. So a neuron model called "filtering neuron" is introduced. The idea is quite simple and natural : neuron with filtering synapse weight would deal with time-variant patterns well. Though this is not a new idea at all, back propagation algorithm for the network composed of filtering neurons is not introduced yet. The tracking simulation demonstrates the performance of the model, and smart tracking strategy is realized.