$\delta{-}$ GLMB Filter Based on Multiple Model Multiple Hypothesis Tracking

The kinematics of maneuvering multiple targets are generally unknown and time-varying. Using only a single frame of data, the conventional δ-Generalized Labeled Multi-Bernoulli (GLMB) filter has large tracking errors. Although the Multiple Hypothesis Tracking (MHT) algorithm utilizes multi-frame information, the amount of calculation will increase sharply with the number of targets and the complexity of environment. To address this issue, this paper incorporates the Multiple Model (MM) and MHT algorithm into δ-GLMB filter. MM is used to generate multiple hypothesis of δ-GLMB target components, and these hypotheses are enriched by association with measurements data on the level of multiple target component. In this way, multi- hypothesis information of each component is transmitted among successive multiple frames. Due to the use of multi-frame information, the performance of δ-GLMB filter for maneuvering targets in complex scenarios is effectively improved. At the same time, the computational complexity can be reduced by forming multiple hypothesis on the level of target components. Simulation results show that the proposed MM-MHT -δ-GLMB algorithm can effectively track targets in multi-target complex motion scenario and has better multi-target tracking accuracy than the conventional single model δ-GLMB filter.