Weighted state combination based multi-sensor fusion for extended object tracking

Distributed multi-sensor fusion tracking has garnered significant attention in recent years, primarily attributed to its remarkable enhancement in both tracking accuracy and field-of-view (FoV) coverage for object tracking tasks. However, in extended object tracking (EOT) applications, when extended objects are close to each other and/or positioned at the boundary of a local sensor's FoV, they are prone to partial occlusion and/or truncation. This phenomenon is liable to induce incomplete measurement data causing the degradation in tracking accuracy. Furthermore, the accuracy of fusion tracking is inherently influenced by the detection capabilities of different sensors, as well as the spatiotemporal distance along with geometric relationship between objects and sensors. To address these challenges, we first introduce partial pseudo-measurements to compensate the incomplete measurements for the truncated extended object (TEO) by using measurement mean and second central moment techniques under the framework of random matrices (RM) model. Then, we formulate fusion weights by integrating localization noise and shape noise, which are further leveraged to fuse estimations for the object's kinematic states and extent states, respectively. Especially, we propose a state combination method to select the optimal combinations between the kinematic and extent states for fusing estimations, which can discover the latent states and enhance the robustness in complex and cluttered environments. Finally, the simulation results demonstrate that the proposed algorithm exhibits superior performance in the tracking scenarios with occluded and/or truncated extended objects.