A Method of Initializing State Parameters of UAV Based on Visual-Inertial Fusion

Abstract

The invention discloses a method for initializing UAV state parameters based on visual inertial fusion. Aiming at the pure visual initialization stage, the feature points that will enter the final global optimization are screened, the feature points in the middle of the image are retained, and the feature points near the edge of the image are selected. The feature points are not optimized, the number of feature points to be optimized and the calculation scale of this stage are reduced, the tracking accuracy of the retained feature points is high, and the change in system accuracy is acceptable. For the visual-inertial calibration stage, the velocity vector of each initial key frame is no longer estimated at this stage, only the gravity vector and scale are estimated, and the 34-dimensional equation of the original algorithm is reduced to a 16-dimensional equation, which reduces the pressure of solving the equation and reduces initialization operation hours. The present invention adds a simple closed solution model before the optimization iteration to reduce the number of iterations and greatly reduce the calculation time.

Description

Technical field [0001] The present invention belongs to the UAV state estimation algorithm design technology field, specifically relates to a UAV state parameter initialization method based on visual inertia fusion. Background [0002] With the rapid development of drone technology in recent years, related technologies for estimating their posture are also developing at the same time. Among them, the typical technique is to establish a joint equation by fusing the camera data with the inertial measurement unit data, solve the initial parameters, and continuously iterate through the back-end solver to achieve accurate pose estimation. Therefore, parameter initialization methods with short time and high accuracy are particularly important in the process of self-state estimation, and have become a hot issue. [0003] At present, the initial state parameters that are accurate are generally divided into two categories: closed solutions and optimized iterative algorithms. The closed s...

AI Technical Summary

Problems solved by technology

There is an existing initialization method using linear estimation, which ignores the effect of gyroscope bias, and does not construct a sensor noise model. Using ORB-SLAM for reference, it solves the gyroscope bias, scale, velocity and gravity step by step, and at the same time Considering that the accelerometer bias has little influence on the acceleration of gravity, the influence of the accelerometer bias is ignored in the initialization process, and the previous initialization algorithm has been improved, and better results have been achieved in terms of accuracy and robustness.

In the initialization process, the IMU pre-integration method is not used, but the visually estimated pose is differentiated to obtain the angular velocity and acceleration, and compared with the IMU measurement value, the error is constructed, and data alignment in the frequency domain is proposed. But at the same time, there is also the need to test whether the function to be differentiated can be differentially calculated

Images