The Research And Application Of Filerting Algorithm In The Autonomous Navigation

As the development of the economy, society, science and technology, the means of traffic become convient and fast, and there explodes the speed of transport and the need for travel. Since the extension of the velocity, frequency and range of human activity, the demand for navigation and location of human beings has never been more intense. Once we conclude technology of navigation and position which have developed for thousands of years, a principle could be reached that the technology is undergoing such process that from the land to sea, from2dimensions to3dimensions, from outdoors to indoors, from slow to fast, from fuzzy to accuracy, from manual to autonomous. During the permeation of high tech through the territory of navigation and location, it is the automation that becomes one feature of navigation and location. And one important reason for the intelligence of navigation and location is the application of mathematic tools—filtering. Therefore, this paper aims the application of the filtering in the autonomous navigation to carry out the research work.Autonomous navigation defines the core of information processing by the Kalman Filter and its derivatives, such as Extended Kalman Filter, Unscented Kalman Filter, Cubature Kalman Filter Robust Filter or Particle Filter. The condition for the validation of Kalman Filter is linearity of the system and the state noise as well as the observation noise must be Gaussian white noise. However, the mentioned condition is difficult to be met. Therefore, the derivatives came into being in order to solve the compatibility of the Kalman Filter and the non-linearity, non-stationary and non-Gaussian of the system.The GPS/INS integrated navigation system is first considered in the paper. Although the integrated navigation system needs the external assistance of Global Positioning System (GPS), the Strapdown Inertial Navigation System (SINS) autonomous navigation is discussed at the GPS outage. For the reliability and accuracy of the autonomous navigation system, the fusion of information from different sensors becomes very important when it comes to the integration of navigation. So, from the above aspect, it can be discovered that the integrated navigation sacrifice the efficiency for the reliability and accuracy. That is why the theory of statistics learning is introduced to enhance the reliability and accuracy of the integrated navigation system. Based on the meticulous inspecting and concluding of the state of art of the research and application of autonomous navigation and filtering, this paper proposes an algorithm for the loose-coupled GPS/SINS integrated navigation system which is able to keep fairly high accuracy at the GPS outage.Also, an algorithm named dual adaptive UKF is designated for the initial alignment on moving basis of which the noise is non-Gaussian and statistic principled unidentified. This algorithm is able to be adaptive enough to scale the covariance of the state noise and observation disturbance to reduce the effect from the change of the covariance.At last, as the indoor navigation in atrium space or long span space structures such as libraries and museums, an algorithm is introduced, which is a low-cost method combining the magnetic and inertial navigation based on smart phones, considering the action patten and gait principle of the users. This algorithm can support the electronic guide in libraries and museums. The main content and contribution are as follows.The accuracy of GPS is often combined with the reliability of INS to accomplish navigation. The simulated annealing optimized SVM is introduced to the GPS/SINS integrated navigation based on the output correction. And the BP-ANN (Backwards Propagation Artificial Neural Networks) and the genetic wavelets NN are briefed. This paper proposes a new robot to accomplish the filter and fusion of GPS and SINS. The state formulation of integrated navigation is first order error formulation due to the adoption of the indirect filtering. Therefore, the KF could gain satisfactory filtering results when the GPS and SINS are both available. And, simulated annealing optimized Support Vector Machine (SVM) is adopted to train the INS data when GPS outages. The algorithm, simulated annealing, is employed to find the best combination of the parameter of the penalty function and kernel function which is key part of SVM used for the training at GPS outage. Therefore, the integration navigation could retain almost as precise as the GPS. In the simulation phase, the optimization effect of proposed algorithm is underlined, as well as the regression effect of SVM training. Moreover, the training effect is proved better after the comparison to the BP-ANN and genetic wavelets NN.A research was carried out on the principle of transfer alignment under the large misalignment situations. And the principle and process of UKF, which is centered by the Unscented Transform and constructed by the Kalman Filter, and other filter algorithms, such as EKF, CKF, PF and Sage-Husa filter are introduced. Besides, error formulations of transfer alignment on moving base are established. In the initial alignment process of the moving base of Strapdown Inertial Navigation System (SINS), model noise and observation interference will affect the estimate of UKF, and then cause inaccuracy in the alignment results. In this paper, the improved UKF algorithm is called dual adaptive UKF. It could not only stabilize model noise by the method of resizing noise, but also inhibit observation interference by monitoring the trace of the adaptive matrix based on innovation characteristics and amending the adaptive matrix in time. In the simulation phase, the proposed algorithm is compared to other filter algorithms, EKF, PF, CKF, Sage-Husa included, in the transfer alignment environment on moving base, both on alignment accuracy and alignment speed. Simulation results show that the dual adaptive UKF algorithm is very sensitive to model noise and observation interference. Also, it has strong robustness which could stabilize the model noise quickly.The pedestrian dead reckoning and localization in atrium space and large-span space has been a problem for the public service departments such as library and museums. This paper focuses on the localization by the smart phone. Further more, the proposed method is independent of any localization algorithm employing RSS. The unique equipment utilized is inertial element and magnetometer embedded in the smart phone. After the summary of the principle of pedestrian gait, this paper proposes a gait recognition algorithm based on smart phone. Then, this paper proposed some adaption to strengthen the step size estimation and direction reckon after the analysis of the fault of current pedestrian reckoning. In the end, taking advantage of the action pattern in certain circumstance, the pedestrian reckoning mainly directed by the low cost MEMS, could be corrected by the geomagnetism which is embedded with the particle filtering. The conclusion could be reached that this method can be easily implemented on the smart phone and is portable, less complex, effective, and low cost.