Research Of Low Cost Viaduct Identification Technology Based On Dynamic Bayesian Network

In a restricted environment,the signals of global positioning system(GPS)satellite are occluded,hence,the applications based on traditional GPS positioning are greatly limited.In this essay,the viaduct is set as an example in urban environment to study the satellite elevation navigation technology in a restricted environment.Not only is the driving problems of drivers solved,but also achieves both low-cost and reliability of autonomous vehicle navigation,possessing profound significance and promising prospect in the field of location service and intelligent transportation system.The essay consists of the following contents:1.Based on the fundamental principle of GPS positioning,the error variance of positioning is analyzed,the result indicates that the accuracy of GPS positioning is related to the number of visible satellites and the geometric distribution of satellite,explaining the fundamental cause of poor elevation accuracy.The advantages of dynamic Bayesian networks(DBNs)for modeling dynamic time-varying stochastic processes with both feature and time series correlations are discussed and two basic morphological expressions of Hidden Markov Model(HMM)and Kalman Filter Model(KF)are analyzed in detail.Moreover,the methods of structure learning and parameter learning in DBN model are discussed further and different process of solution for parameter learning with observation data and incomplete observation data is also deduced.2.Aiming at the problem that the GPS signal in the elevated environment is weak and the traditional car navigation cannot be recognized by itself,an improved viaduct recognition model based on multi-state transition DBN is designed.Combining with the actual scene of the vehicle traveling on the viaduct,the relevant state variables and interdependence are extracted,then the parameters such as state transition distribution and observation probability distribution among the state nodes are learned,including the following aspects.First,learn model of vehicle traveling on the viaduct with binomial logistic regression and obtain the parameters of the model in the descending gradient method;then describe the transition process of the vehicle through the weighted moving average model to eliminate the random fluctuations in the prediction;finally,adopt the pseudorange residual mean square error sum to derive the observation model of the recognition process to carry out the judgment of the driving condition of the vehicle and the estimation of the driving height,relizing the low-cost viaduct identification based on GPS data.3.The probabilistic reasoning of dynamic Bayesian networks is studied,and the posterior probability density function of the target state estimation of dynamic systems is derived.For the nonlinear and high dynamics of the viaduct decision model,two nonlinear filtering methods,unscented Kalman filter(UKF)and particle filter(PF),are analyzed.The simulation experiments of different process noise variance and observed noise variance are compared and analyzed and the results show that PF has better filtering performance and robustness,so PF is more suitable for state estimation of nonlinear and highly dynamic systems.The ability of different resampling algorithms to improve the particle depletion problem in particle filter algorithm is discussed and contrasted.Combined with the actual situation of the viaduct recognition model,an improved particle filter algorithm is proposed to reduce the correlation between particles.Besides,it increases a fault tolerance.In this way the state of particle distribution is more real,so the performance and stability of filtering are improved.4.The correctness and rationality of the viaduct recognition model based on complex dynamic Bayesian network and the improved particle filter inference algorithm are verified by vehicle experiments.Firstly,in the single-frequency pure GPS data recognition model,the binomial logistic learning is carried out with the visibility rate of satellite and the probability of the vehicle traveling at the previous moment,so that the improved PF algorithm is adopted to judge the driving condition and the driving height of vehicle.The results indicate that the method can be well recognized and can maintain stable recognition efficiency in different scenarios,reaching over 96%.The estimation accuracy of vehicle travel height is 3.7 meters on average,which is enough for most viaduct identification.This result proves the validity and universality of the decision network model and the improved particle filter algorithm,and is suitable for low-cost car navigation systems.5.A delay bridge identification method based on low-cost barometer is proposed to improve the delay of the vehicle during the ramp.Therefore,the decision model of the viaduct based on complex dynamic Bayesian network is improved to achieve continuous,accurate and real-time positioning.Since the current smartphones and most mainstream mobile phones are equipped with barometers,the smartphone barometric data acquisition software is designed.This method does not require high accuracy of the barometer,and provides an inexpensive and feasible car navigation system.In a word,the viaduct solution has broad prospects for promotion.