Research Of Key Technology Of Driving Behavior Recognition Based On Information Fusion

With the development of social economy, motor vehicles have been increasingly growed, and more and more attention has been paid to the personal security .According to the domestic and foreign research,the rapid develpoment of modern vehicle manufacturing technology makes the proportion of the vehicle's safety factors in traffic accidents much smaller, and personal factors of the drivers became the first factor of traffic accidents, and the main factor that caused traffic accidents .Therefore, many scholars have carried out a series of research of automobile active safety system based on human factor in recent years.But a lot of research work is confined to one single driving behavior or relationship among isolated human-vehicle-road,without fully considering potential danger to vehicles and road traffic brought about by driving behavior ,and how to prevent traffic accidents effectively based on comprehensive environmental information of drivers- vehicles-road. Therefore,In this paper,it applied the method of virtual sensors, make full use of the sensors installed on the vehicles ,form a human - vehicle - road closed-loop system of multi-sensor data acquisition and preprocessing system through the installed car visual system.Research and develop driving behavior recognition model based on multi-sensor information and algorithm, identified driver's driving behavior in time,and generally assessed the security of such driving behaviour according to the driving behaviour identified and environment information,gived the driver the appropriate alarming information according to the results of the assessmen,then the information was transmitted to the assistant drive system on the vehicle,which takes relevant prevention and control measures.It firstly introduces driving behavior and the research achievements and problems of automobile active safety both at home and abroad. According to recent designing ideas of human-centered active safety system,The paper proposed the idea of developing active automotive safety system based ondriving behavior,and researched the key technology of driving behavior recognition.The paper used the multi-index synthetic evaluation method—AHP to do numerical analysis of main composition factor of driving behaviour and initial decision-making weight coefficient and identified vector that solved the common main composition factor of driving behaviour.Established decision-making weight and intial identify vector of each information in the multi-sensor information in a certain driving behaviour,provided basical theoretical analysis work for driving behavior recognition model based on multi-sensor information and algorithm,that used BP neural network and D-S evidence theory.Meanwhile. At the same time to the characteristics of driving process information and the three phase in driving process: perception, decision-making and operation, studied the relationship between the three of the human-vehicle-road closed-loop system and the influence brought to the car.Aiming at the status that various sensors constantly being installed in the car to get various safe driving purposes, and in order not to increase the number of sensor of car, the paper comed up with using the technology of virtual sensor to solve the problem of information sharing among multiple sensors, adopted the technology of software and algorithm to get information of virtual sensor driver ,which must get for driving behavior recognition from the existing combination of single or several sensors in the car, solved the large information driving behavior identification needed and the information sharing.In this paper,the installed vehicle vision recognition system was specially designed,and the vision sensor processing recognition unit whict was composed by CCD sensor and DSP,then thoroughiy studied the composition,calibration and algorithms of the vehicle vision recognition system.Researched the mathematica model and algorithm based on the location of he installed vehicle vision recognition system.The paper analyzed the advantages and disadvantages of commn method of the multi-sensor information fusion technology, according to the state of driving behavior instantaneous changeful and subjectivity strong's characteristics, proposd to use the model and algorithm based on BP neural network and the D-S evidence theory combiniation of combined-type multi-sensor information fusion. And in the light of D-S evidence theory decision-making recognition application, combined with the specific application of driving behavior identification, the paper proposed the key evidence method to improve traditional D-S evidence theory decision-making algorithm, made the improved D-S evidence theory is more suitable for driving behavior identifying decisions,according to using the combination modle of BP neural network and D-S evidence theory, make the various multisensor information meaning clear and transparent in the whole multisensor information fusion modle.Finally, in order to meeting the demand of test, the paper designed the software and hardware in the loop simulation platform of driving behavior identification based on ARM + DSP dual-core .Respectively on simulation platform involves hardware design,software design based on Linux, communication interface, agreement and interface etc gived a detailed design idea and method .Based on the completion design of hardware and software simulation platform,at the same time used the tools of the Advanced Road & Advanced Driver and DYNAanimation of ve-DYNA vehical dynamics simulation software to designed simulation driving behavior data extraction model and method,extraction of driving behavior numerical value offline parameters which was needed recognition,testing driving behavor identification method, algorithm and the reliability of software and hardware simulation platform. Test results showde that the thesis based on combined-type multi-sensor information fusion driving behavior identification model and algorithm can effectively solve the real-time driving behavior identification's key technical problem.Mealwhile for software and hardware in the loop simulation platform for the various peripheral communication interface and human-computer interface was tested as well,the tests indicated that the designed simulation platform communication interface and the agreement was reliable,which can obtain effectively related peripheral communication parameters for further research and development based on the car driving behavior active safety systems development laid a foundation.Finally the contents of this paper and results were summarized. Especially for the next step of this paper's research route, ideas and methods ,that is ,on the basis of current study,proposed establishment of hidden Markov chain is used to establish driving state transition prediction mode, constructing driving behavior real-time identification-forecast- evaluation of comprehensive driving behavior identifying prediction model and algorithm.