Probabilistic Flight Envelope Based Joint Tracking and Classification Using Kinematic Measurements

For direct classification using kinematic measurements, existing flight envelope (FE) based approaches may be insufficient. This is because the FE primarily provides region information of the kinematic variables for targets in a class, whereas the regions for different classes often partially overlap, leading to potential ambiguities in classification. To address this, this letter proposes utilizing a probabilistic FE which provides not only the region but also the probability density function (pdf) of the kinematic variables. To fully integrate both types of information, a constant position with a velocity-acceleration input (CP-VAI) model is proposed. In this model, each input is confined within the envelope region, and the mean-square-error matrix of the input, which contains the first and second moments of the pdf, is used as the covariance of the process noise. Based on this model, a joint tracking and classification approach is proposed. In this approach, class-based tracking can be performed efficiently using the best-model-augmentation algorithm, where each candidate model is assumed to be a CP-VAI model and a class-based model-set-adaptation method is developed to select the best model for tracking. For Bayesian classification, the likelihood function of the best model can be used as that of the class. The proposed approach naturally integrates the probabilistic FE through the CP-VAI model and has an analytical form. Experimental results demonstrate the effectiveness of the proposed approach.