To improve the capability of detecting weak moving target with unknown statistical feature

a kind of track-before-detect (TBD) algorithm implemented by cost-reference particle filter (CRPF) is proposed (CRPF-TBD).Firstly

a discrete variable mimicking the existence and absence of a target is added to the state vector of CRPF.Particularly

besides transform probability matrix

the transformation of the discrete variables in the CRPF-TBD also depends on the correlative coefficient of successive measurements.Secondly

the existence probability at each frame is calculated from the output discrete variables.Finally

a binary test statistic is constructed from the existence probabilities.Target is declared to be present if the test statistic exceeds a given threshold.Simulation experiments of over-the-horizon radar (OTHR) target detection show that

in the case of given statistical feature

the detection performance of the proposed method is comparable to that of the traditional particle filter (PF) based likelihood ratio algorithm and existence probability detection method

and improved by more than 2dB when the statistical feature is unknown.The proposed method could be applied to surveillance in complex environment

such as radar and sonar.