Fusion framework for moving-object classification

Perceiving the environment is a fundamental task for Advance Driver Assistant Systems. While simultaneous localization and mapping represents the static part of the environment, detection and tracking of moving objects aims at identifying the dynamic part. Knowing the class of the moving objects surrounding the vehicle is a very useful information to correctly reason, decide and act according to each class of object, e.g. car, truck, pedestrian, bike, etc. Active and passive sensors provide useful information to classify certain kind of objects, but perform poorly for others. In this paper we present a generic fusion framework based on Dempster-Shafer theory to represent and combine evidence from several sources. We apply the proposed method to the problem of moving object classification. The method combines information from several lists of moving objects provided by different sensor-based object detectors. The fusion approach includes uncertainty from the reliability of the sensors and their precision to classify specific types of objects. The proposed approach takes into account the instantaneous information at current time and combines it with fused information from previous times. Several experiments were conducted in highway and urban scenarios using a vehicle demonstrator from the interactIVe European project. The obtained results show improvements in the combined classification compared with individual class hypothesis from the individual detector modules.