Attitude and position estimation method based on vision and inertia information

Abstract

The invention discloses an attitude and position estimation method based on vision and inertia information. According to the method, by fusing vision and inertia information of an image, motion state information is saved and updated by virtue of an expanded kalman filter so as to calculate the current accurate attitude and position of equipment, and the motion trail of the equipment within a certain duration of time can be obtained. According to the estimation method disclosed by the invention, the information of the image and a sensor can be flexibly utilized so as to achieve good robustness through mutual complementation of the information, and a situation of tracking losing is avoided. The expanded kalman filter adopted by the method is used for calculating based on an iterative form, and compared with a method of batched calculation, the estimation method can be used for calculating without acquiring all observed quantities, is relatively low in calculated quantity, adaptive to a condition of relatively few calculation resources at terminal equipment and capable of preferably achieving a requirement on real-time performance.

Description

Technical field [0001] The invention belongs to the technical field of device posture and position tracking, and specifically relates to a posture and position estimation method based on vision and inertial information. Background technique [0002] At present, there are two main positioning methods commonly used on terminal equipment: one is to use inertial units, that is, to obtain the device position through the data of gyroscope, gravity accelerometer, and accelerometer; the other is to process the visual information collected by the camera. The posture information of the device. Due to the limitation of sensor accuracy, the former method is prone to accumulate errors and cause the estimated drift of the position, which is more obvious when the movement speed of the device is small, especially the zero speed detection is still a difficult problem. The second method is based on visual information and uses the multi-view geometry method to recover the position of the visual te...

AI Technical Summary

Problems solved by technology

Due to the limitation of sensor accuracy, the former method is easy to accumulate errors, resulting in position estimation drift, which is more obvious when the movement speed of the equipment is small, especially zero speed detection is still a problem

The second method is based on visual information, and uses multi-view geometry to restore the position of the visual terminal relative to the scene. This type of method does not have the problem of error accumulation, but limited by visual information, this type of method cannot obtain the scene Mesoscale, which can only estimate the relative orientation, and requires the terminal’s moving speed to be small, otherwise, large calculation deviations and even position loss will occur due to the lack of matching points between image frames

The main advantage of this type of method is to use the inertial sensor to predict the movement of the device and speed up the feature matching of the image, but it cannot greatly improve the final accuracy.

Images