Abstract:

A novel estimate-to-truth assignment algorithm based on the optimal assignment is proposed for performance evaluation of multi-object filter. The corresponding covariance matrices of the estimates are used to detect a part of false estimates and missing objects based on the standard elliptical gating technique. Then, the residual estimate and object set are assigned based on the optimal assignment. Since the Mahalanobis distance and the standard elliptical gating solution are used in the process of assignment, the novel assignment algorithm avoids scaling problems and it is unrelated with the concrete applications. To evaluate the multi-object filters, a multi-object missdistance is generalized. Numerical analyses verify the validity of the proposed assignment algorithm and the defined measures are shown to be effective in capturing the accuracy of the multi-object filters.