Research On Vehicle Recognition Algorithm Based On Evidence Theory

Multi-source vehicle recognition is the vital technology of intelligent transportation.It is the key problem of automatic classification of different vehicle types to process the uncertainty,inaccuracy,imperfection of information sources properly.In information fusion algorithm,DS(Dempster—Shafer)evidence theory can deal with the uncertain information effectively,express and combine them reasonably,for which DS evidence theory has aroused great concern among scholars.There are some difficulties in both of theories and applications about how to deal with the conflict evidence and how to turn the practical problems into basic probability assignment function in the reasoning.Based on DS evidence theory and its application in the vehicle recognition,this paper has made some deep research on evidence theory.The follows are the main study contents:1.This paper introduces the multi-sensor information of principles,fusion and structure model,and makes the induction and summarization for target identification.2.The paper makes some analysis about the principles on evidence reasoning,its development and combination rules.It also makes the analysis about the problem from evidence theory utilized in the application.This paper makes the analysis about evidence conflict origin.At present the primary solutions for evidence conflict are inducted and summarized in this paper,including open frame of discernment,amend Dempster combinationrules and original evidence sources.The previous improvement method is an effective method of evidence conflict in some degree.But there are still some shortcomings of high confidence,namely high uncertainty,and low recognition rate.Aiming at these drawbacks,a new improvement method is proposed on the thought of predecessors in this paper,based on principal component evidence.The simulation result shows that the improvement method can make effective fusion among the conflict evidences and optimize recognition rate.3.Image feature extraction is the key problem in image target recognition.Because of different principles on visible light imaging and infrared imaging,the methods of feature extraction have to be different.An optimal method-wavelet moment feature extraction has been proposed,aiming at visible light feature extraction.The result proves that the wavelet moment can keep translation,rotation and invariable.Finally two single sensors are utilized to identify the vehicles and the identification result can support the later chapter as experimental data.4.It is quite hard to obtain the basic probability assignment function for the application of DS reasoning theory,for which this paper proposes its structure method.In accordance with the studies above,the simulation experiments are established for vehicle identification on the basis of evidence theory.The result testifies the availability of multi-sensor information fusion in target identification.It is also can be showed that after adjusting the fusion rules the frame of discernment confidence can be cut down and the correct rate of vehicle identification and classification can be improved.