A Data Association Approach Based on Multi-rules in VorSLAM

To solve the problem that the data association (DA) algorithm based only on Mahalanobis distance cannot ensure to produce correct results, this paper proposes a data association approach based on multi-rules, which has utilized the characteristic of the hybrid metric map representation, adopted in VorSLAM. This approach consists of the rule of local searching, the rule of unidirectivity existing in the corner's observation, the rule of Mahalanobis distance and the rule of shape matching. These rules execute in turn in each data association circle. The former two rules decrease the searching space of the data association and avoid a sort of wrong data association. The rule of Mahalanobis distance can produce a set of potential data association hypothese according to the location information expressed in the feature's parameters. Utilizing the characteristic that in VorSLAM each local map describes the local environment's contour of its corresponding feature, the rule of shape matching finally recognizes the correct data association hypothesis. The data association based on multi-rules has systematically and reliably solved the data association problem in VorSLAM. The efficiency of the proposed approach is verified in two experiments carried out in indoor environments.