Hybrid Kalman Particle Filter for Ground Target Tracking

We propose to continue our Phase I efforts by demonstrating the class of hybrid Kalman particle filters (HKPF) for tracking ground targets on road in Phase II. In the proposed formulation, the road network is represented with one-dimensional (1D) models. This 1D modeling simplifies the target kinematics considerably and reduces the size of the target state space by up to ½, which directly translates into the computation and memory savings “desperately” needed by particle filters. It also allows for explicit incorporation of the road information and interactions with other targets into the filters. Together with features aiding from HRRR and/or SAR (simulated from FATSO in Phase II), it provides additional means for better report-to-track association and tracking of targets across an intersection (move-stop-move with or without turn). Furthermore, the measurement nonlinearity arising from the 1D modeling of roads and target feature aiding is well handled by the proposed nonlinear filters. In Phase II, the computational algorithms developed in Phase I will be implemented into well-structured software modules toward a software development toolkit as part of the Phase II deliverable and a “marketable” product to pursue after Phase II. At the same time, we will decompose the nonlinear tracking filters into pipelined parallel structures and implement it in a Mercury RACE++ Series multiprocessor platform for demo.