On the vehicle recognition based on improved evidence reasoning

Vehicle recognition has become a hotpot in the researches on intelligent traffic management system. In the vehicle identification, the collision information is fused using the combination rule of Dempster-Shafer evidence, and the irregular results will be occurred. It is discussed that a new calculation method of weight coefficient in this thesis. In this method, the evidence principal elements for each group can be determined by analyzing the principal component of evidence. Then the evidence compatibility and credibility are obtained. Further, the weight coefficient of evidence can be determined through all above parameters. An evidence conflict measure method is put forwards in this paper. In this method, the conflict value can be calculated, then weight coefficient is normalized, the evidence source is modified by normalization weight coefficient. Further, ER evidence is used to identify the target. In the simulation progress, the vehicle type on realistic road is regarded as the recognition example, and it can be utilized to compare this method with others. It can be proved that this method is more effective than others on fusing the evidence with high conflict and reducing the computational complexity. The accuracy of target recognition has improved by 20%.