A Modified Visual Compass Attitude Estimation Method
Bo Han · Chuangan jishu xuebao · 2011
The visual compass proposed by Montiel et al.is an attitude estimation method based on EKF-SLAM that using camera as the only sensor.When camera rotation is unsmooth,a priori system noise covariance in motion model should be set to a large value,but it results in high computational cost for matching and high rate of matching error occurrence.Focusing on those problems,a modified visual compass attitude estimation method is presented.First,a multi-resolution landmark selection strategy is used for new landmark initialization.Then,an active search/match layer by layer algorithm is used to reduce both the computational cost and the rate of error occurrence.Finally,a RANSAC algorithm sampling 2 matched points is used to remove the outliers in matching result.The modified visual compass attitude estimation method is applied to the outdoor video,and the experimental results shows that the modified method has advantage of less computational cost and improved accuracy of attitude estimation.