An Integrated Navigation System Of GNSS/VO Based On Adaptive Kalman Filtering

The navigation technology with timely and precisely proformance is one of a key technologies to intelligent vehicles,and it is the premise of the vehicle's auto-movement.Satellite's acquisition is the premise of tacking and positioning,and the time of acquisition is an improtant parameter to GNSS receiver.Since the satellite signal may be occluded,the navigation data provide only by satellite is not meet with the stably positioning requirement of intelligent vehicles.The camera is widely equipped in intelligent vehicles due to its advantages of low price,high resolution positioning information and low operation requirements.Therefore,the monocular VO and GNSS combination method is studied in this paper.Firstly,this paper have proposed a PFA algorithm for Bei Dou software receiver based on coherent downsampling aim to solve the problem of large computation load of traditional PFA algorithm.A complex down-conversion module and a coherent down-sampling module have been added to the traditional PFA architecture.Compared with the traditional PFA algorithm,the time required to acquire the satellite was reduced by at least 80% and,although the signal-to-noise ratio(SNR)lossed as the number of coherent integration points M increased,the loss in SNR did not exceed 0.5d B when M was less than 500.Then,combined the advantages of FAST and SIFT in feature detection,this paper have proposed a modified FAST algorithm.32-dimension descriptor has been added to the modified FAST algorithm.According to the characteristics of the vehicle's odometer,this paper have proposed a two-step method to remove mismatched feature,the first step is to remove visible mismatched feature that can been find by naked eye,the second step is to remove the mismatched feature that are located on moving objects.Compared with SIFT algorithm,the time used on feature detection has decreased by 76.46%.The number of correctly matched feature pairs was more than 8 that were met with the requirements of the visual odometer posture calculation.Finally,based on the plane vehicle motion model,this paper proposed a plane constraint-based VO solution model.By this model,Whether the monocular VO's scale have been repaired by RTK or single-point GPS,the monocular VO had been runing along a straight line for 30 s,the maximum position error was 0.2652 m,and the maximum yaw error was 3.9672°.According to the adaptive Kalman filter,the position error and yaw error of the two sensors has been used as the input of the combined system,and the VO&GNSS loose combination model have been proposed.Whether the monocular VO is conbined with RTK positioning data or with single point GPS positioning data,the VO&GNSS combined system had been runing along a straight line for 30 s,the maximum position error was 0.1403 m,and the maximum yaw error was 2.5916°.