A TRACK INITIATION METHOD BASED ON THE COMBINATION OF KALMAN FILTERING AND THE TRANSFORMER MODEL

Aiming at the problem of track initiation of multiple moving targets in a strong clutter environment, this paper proposed a track initiation method based on the combination of Kalman filtering and the Transformer model. Firstly, Kalman filtering is used to filter the radar measurement data to generate a temporary track set containing real and false tracks, then the temporary track set is mapped into a set of spatio-temporal vectors that can reflect the track characteristics, and finally, the spatio-temporal vectors are input into the Transformer model for classification, so as to get real tracks to complete the track initiation. The Transformer model uses a position-coding layer and self-attention mechanism so that the model can directly capture the dependencies in the sequence and extract the motion characteristics of the target efficiently. The experimental results show that compared with the traditional sequential processing methods and batch processing methods, this method can not only complete the track initiation in a shorter time but also has a higher track initiation rate and a lower error rate in the strong clutter environment, which is of great significance to target tracking.