A 3D Multi-Model Tracking Algorithm Based on Sequential Filtering

‍ ‍Multi-model algorithms can more effectively deal with the diversity and uncertainty of target maneuvering movements owing to their ability to handle different scenarios and changes， as single-model tracking algorithms can no longer meet the tracking requirements of high maneuvering targets in three-dimensional （3D） maneuvering target tracking. Compared to two-dimensional maneuvering target tracking， three-dimensional high maneuvering target tracking requires more models to form a model set for precise tracking owing to its complex spatial dynamic characteristics. This situation leads to the complex design of classic interactive multi-model algorithms and the excessive consumption of computing resources. However， traditional multi-model algorithms have poor tracking performance because they do not consider the interaction between models. This study proposes a new multi-model tracking architecture based on sequential filtering to address the above issues. This architecture combines sequential filtering with multi-model algorithms， using the fusion result of multiple models from the previous moment as the input for each model filtering in the next moment. This process greatly simplifies the complex interaction implementation process in interactive multi-model algorithms. Concurrently， for the problem of model set combination， the 3D motion is decoupled into a combination of several two-dimensional and one-dimensional motion models， forming a new 3D space maneuvering target-tracking model set. The tracking comparison results of simulation and measured data show that compared to traditional multi-model algorithms， the proposed multi-model algorithm based on sequential filtering correction demonstrates significantly improved tracking accuracy， comparable to interactive multi-model algorithms； however， the proposed algorithm is far better than interactive multi-model algorithms in terms of computation time， significantly reducing the demand for computing resources.