Research On Track-before-Detect Algorithm Based On Particle Filters

Dim small target detection and tracking is the key technology of the infrared surveillance system, precise guidance system, and satellite remote sensing system. As the long distance propagation and the influence of strong noise, the signal-to-noise ratio (SNR) of the received data is very low. In this case, the traditional detection and tracking methods which apply a threshold to a single data frame can not meet the requirements. An alternative approach, called Track-Before-Detect (TBD), is to supply the tracker with all the sensor data without applying a threshold. And detection is declared after target energy accumulated by time. This can improve detection performance and allow the tracker to track low SNR targets.The particle filter (PF) based TBD is very powerful and versatile. However, a large number of random samples are needed for a good performance, which cause a large amount of computation. This is disbenefit for engineering realization. This paper mainly studies PF based TBD algorithms, and applies several techniques to reduce the computation burden of the algorithm.Firstly, by analyzing the model of infrared dim small target, we proposed a new TBD algorithm based on marginalized particle filter (TBD-MPF). In the algorithm, velocity of the target state vector which appears linear Gaussian is marginalized out and estimated using an optimal Kalman filter, while the position and intensity of the state which enter the measurement equation nonlinearly are applied with particle filter. With the reduction of the state dimension, the computation load is significantly reduced.Secondly, by replacing the conventional Monte-Carlo (MC) points used in particle filter by the Quasi-Monte Carlo (QMC) points, we develop an improved algorithm: the QMC based Gaussian particle filter (QMC-GPF). Because the integration error from N samples drop as N-1 with QMC, which is much better than N-1/2 with MC, the QMC-GPF algorithm can save a lot of computation while keeping same precision.Finally, based on the QMC-GPF algorithm, a new TBD algorithm is proposed. The convergence characteristic of the covariance matrix of the posterior densities propagated in the QMC-GPF is used to determine whether it is the true target. The algorithm is tested on both simulation and real data and is shown to be able of performing sufficiently well for the target whose SNR is above 3dB.