Learning to Track Multiple Radar Targets with Long Short-Term Memory Networks

Figure 1

Overview of our architecture. In a time step , the architecture is composed of two parts as target motion module and target association module (see Section 3.2 for details). The input can be expressed as the following: at the current moment , the target motion state , the hidden state , and the probability vector (see below) come from the learning of the previous moment . After the processing of our architecture, the output is the estimated target motion state , the hidden state , and the probability vector of the next moment .