Research On Camera Pose Estimation And Target Location Based On Vision And Inertial Device

The problem of camera pose estimation is one of the research hotspots in the field of computer vision,and also a key issue in the research of target positioning.It has great applications in robot navigation,augmented reality and other aspects.However,the actual camera’s posture is affected by factors such as platform motion and vibration,which can cause jitter or sway to cause image shake.It is difficult to accurately and effectively estimate the camera’s posture based on the corresponding point-based pose estimation method.The inertial device attitude measurement technology has a high degree of autonomy,is not easily affected by the environment,and has a wide application space.Therefore,this paper studies the camera pose estimation problem based on the inertial device fusion of visual information,and under the fusion framework of the visual inertia based on loose coupling filtering,a visual inertial fusion model based on line feature parameters is proposed,and the fusion model is applied to vehicle and airborne platforms to estimate the camera attitude and target positioning,and at the same time,the research on ship target detection,ship target feature extraction and sea surface feature under high-altitude airborne conditions is studied,which solves the problem of error accumulation in the attitude estimation process relying solely on inertial devices and ship target detection in high-altitude environment.Good results have been achieved in the actual flying experiment.The main work of this paper is:Firstly,for the error calibration problem of inertial devices,an accelerometer calibration method based on theodolite is proposed.This method uses the acceleration of gravity to generate excitation on the three axes of the accelerometer,and uses Kalman filtering to iteratively converge the calibration coefficients.The experimental results show that this method can converge the calibration coefficient to an ideal value.In terms of gyroscope calibration,without relying on a rotating platform,a method of gyroscope parameter calibration based on a visual sensor is proposed,which uses the feature point coordinates in the camera coordinate system to calibrate the gyroscope scale factor and zero offset parameter.The experimental results show that the method can accurately calibrate the parameters of the gyroscope.Secondly,in view of the problems in the process of camera pose estimation and target positioning with a single sensor,a model of multi-sensor fusion for camera pose estimation and target positioning is proposed.Including the fusion of laser rangefinders to solve the problem of lack of depth information of the vision sensor.In view of the problem that it is more difficult to track and match the feature points when using point features in the visual inertial fusion model based on loose coupling filtering,this paper proposes a posture solution based on line features in the front-end visual sensor.The line feature parameters are used as visual inertia fusion method to fuse and filter the visual output and the posture result of the inertial device.The line feature parameters solved by the visual sensor are used as the state update,and the posture solution of the inertial device is used as the state prediction,and based on this model,estimates the camera posture and target positioning of the vehicle platform and airborne platform.Then,for the problem of camera attitude estimation under the vehicle platform,in the case of the traditional GPS signal losing lock,a research method for vehicle camera attitude estimation based on low-precision inertial devices is introduced.This method uses lowprecision inertial devices combined with vehicle speed.The sensor obtains the accurate attitude of the vehicle on the basis of the vehicle dynamic equations.The experimental results show that the attitude measurement method can accurately estimate the attitude information of the vehicle even under high dynamic conditions.In terms of the fusion of in-vehicle cameras and inertial devices,based on the visual inertial fusion model framework of line feature parameters,a method of fusing the vanishing point parameters as the visual output and the attitude solution of the inertial device is proposed.The experimental results show that the method can effectively solve the camera pose on the basis of suppressing the zero drift of the inertial device.At the same time,in view of the problem of measuring the vanishing point parameters in the visual information,the output of the inertial device is used to assist the measurement of the vanishing point parameters of the visual information.The experimental results show that the vanishing point parameters in the visual information can be accurately measured with the assistance of the inertial device.Then,for the problem of sea surface target detection and feature extraction under the airborne platform,a wake feature recognition method with the assistance of inertial devices is proposed.The wake feature is the most obvious feature of the ship target that can be observed at high altitude.To identify specific ships is an effective method for detecting ship targets at high altitude.Experimental results show that the method can still detect the ship target to be located even when the other characteristics of the ship target are completely invalid at high altitude.In the research of sea surface features,based on the extracted sea and land features,two super pixel-based sea and land segmentation algorithms are introduced.The first is to use the extracted sea and land features to perform multi-feature fusion of sea and land segmentation.Experimental results show that this method can segment sea and land regions,but the segmentation result depends on the initial superpixel segmentation result.The second method is a multi-scale sea and land area segmentation algorithm based on superpixels,which performs multiscale feature statistics on superpixel blocks.Experimental results show that this method can overcome the defects of the first method and achieve better results than the first method.Finally,for the camera pose estimation problem on the airborne platform,based on the visual inertia fusion model framework of line feature parameters,a pose estimation method using sea-sky line parameters as visual output parameters and inertial device fusion is proposed.Under the extended Kalman filtering theory framework,this method combines the sea-sky line observation model and the rigid body rotation model to estimate the attitude of the carrier.By predicting the sea-sky line parameters and its residual covariance to judge whether the sea-sky line is detected correctly,the experimental results show that the method can accurately detect the sea-sky line and estimate the attitude of the carrier,and in the process of detecting the seasky line,a method based on the sea-sky line detection algorithm of inertial measurement and template matching fusion,based on the position and angle of the sea frame of the current frame,combined with the inter-frame inertial measurement data,predicts the next frame of the seasky line area.The experimental results show that this method can reduce the sea-sky line detection area,at the same time can greatly improve the detection accuracy.On the basis of estimating the attitude of the airborne camera,a sea-based ship target positioning model based on the airborne camera attitude is established,and the accuracy of the established positioning model is verified by the actual flying experiment.The experimental results show that even in high-altitude environment,it can locate the target of the ship on the sea.The research of attitude estimation and target positioning technology based on visual information and inertial devices is helpful to solve the problem of error accumulation in the attitude measurement of inertial devices,and it extends the algorithm of visual information extraction and inertial device fusion.It can provide technical support and method reference for research on inertial device navigation and attitude measurement.