Research On The Visual Odometry Aided INS/GPS Integrated Navigation System

The integrated navigation system based on INS/GPS becomes one of the most import navigation technologies in various fields due to their performance complementation.However,this integrated navigation is almost totally dependent on the GPS signal,and the GPS signal is extremely easily blocked or even invalid since artificial or environmental disturbances,which will significantly reduce the positioning accuracy of the integrated navigation system based on INS/GPS.To improve the positioning accuracy of the integrated navigation system,some other aided sensors are needed when the GPS signal is denied.With the rapid development of the computer vision,the visual odometry(VO)which is low price,low power consumption and informative gets more and more attentions quickly.Meanshile,the VO aided INS/GPS integrated navigation system has become a hotspot and a researching direction in the navigation field.How to extract the surrounding environment's information from image sequences obtained by visual sensors is the basis and crucial step of the visual odometry,which is also one of the key technologies in the the VO aided INS/GPS integrated navigation system.What's more,given that the VO aided INS/GPS integrated navigation system has strong nonlinearity,researching on the nonlinear filter of the integrated navigation system is necessary.Thus,this thesis focused on the above points,and the main works were as follows:Three of the most widely used feature detectors,named the Scale Invariant Feature Transform(SIFT),Speeded-Up Robust Feature(SURF)and the Feature from Accelerated Segment Test(FAST),were introduced in detail.Their performance was compared and evaluated by using different image sequences.The results showed that the FAST detector can detect and extract features more efficiently and availably,but it is sensitive to the noise.To solve this problem,an improved feature extraction method was proposed,in which the matched features are refined by the Random Sample Consensus(RANSAC)method.Thus the influence of the noise can be decreased and the robustness of the improved feature extraction method can be enhanced significantly.Based on the improved FAST feature extraction method and the binocular visual odometry method,a new improved binocular visual odometry algorithm was presented in this thesis.In this new algorithm,using the improved FAST feature extraction method to extract features and match them,the location and navigation parameters of the binocular visual system can be estimated by using the high accuracy meatching features and the binocular visual odometry method.And the feasibility and superiority of the presented algorithm were verified by the actual experiments.In the VO aided INS/GPS integrated navigation system,the statistical properties of the system are unknown and the system has strong nonlineatity.Regardless of these,the filter accuracy will be reduced.In this thesis,the Cubature Kalman Filter(CKF)and the Variance-Covariance component Estimation(VCE)method were used to solve the nonlinear problem and to estimate the system noises,respectively.Therefore,we provided a new nonlinear adaptive filter,referred a CKF adaptive filter based on VCE method,to improve the filtering accuracy of the integrated navigation system.At last,we proposed an integrated navigation algorithm ultilizing the CKF adaptive filter based on VCE method,and the above algorithm was applied in the actual vehicle experiments when the GPS is denied.The results showed the effectiveness and superiority of the above integrated navigation algorithm.