Pose Measurement With Multiple-view Space Geometry

Pose parameters,such as the position and orientation,which can reflect the spatial state of targets are core parameters.It has close relationship with the continuous development of aircraft,robot,electronic devices and many other modernized instruments,so that pose parameters measurement has shown essential application value in many industry fields,such as the robot navigation,aerospace,navigation and surgery,etc.Strategies that combine image sequences by shooting and the intrinsic and extrinsic parameters of cameras become the main trend in pose estimation,due to their practicability.In practical applications,how to precisely measure the pose parameter of the target by a known set of data has become an urgent problem to be solved.Existing pose estimation approaches based on photogrammetry could be divided into monocular,binocular and multi-camera vision measurement according to the number of optical devices(cameras)used to tracking target,or be divided into the modeling solution and the intersection measure according to the core algorithm.However,there is a limitation on the precision of monocular measuring results,because only one station could not provide complete information.While the model-based algorithms usually become impractical since they require pre-prepared training template library,which need a large amount of the work.Most of the algorithms need to specify the controlling points on the targets and the models,which are not suitable for many applications.Therefore,in this paper we focus on designing a multi-camera vision algorithm based on intersection measurement.After a comprehensive study of existing methods and the knowledge of photogrammetry and computer vision,we provide a target pose estimation method based on multiple-view space geometry.Our algorithm is a non-contact measure method,which neither need camera calibration beforehand,nor training template library.We need image sequences by shooting from different views and corresponding pose parameters of cameras at the moment of shooting.We can get more precise results by a simple human-computer interaction in computational process.It is suitable for some targets that have simple contour and can extract some key information.In this paper,we spend most of the time on the research and analysis of two key stages in our algorithm,image processing and the optimized pose parameters solving.In the image processing stage we analyze the image sequences and extract the essential information,as well as ensure the corresponding relationship among images.Firstly,we use improved SFM algorithm toestimate the intrinsic and extrinsic parameters of cameras.Then,we associate the image sequences and analyze the spatial target which was filmed.Based on the study of several classical algorithms of line extraction,contour detection and the actual appearance of the objects,we propose and implement an algorithm,which combines line matching and epipolar constraint,to determine the corresponding points.With ensuring the accuracy,we determine the matching information of images simply and fast in order to acquire the 3D information of objects.By solving the pose parameters,which means finding out the position and orientation of targets by using the information extracted from the image sequences,camera pose parameters and the appearance of object,we try to minimize the measurement error.In this paper,we get the optimal solution by using the intersection algorithm of matching points,lines and faces by linear triangulation,trifocal tensor and so on.And we get the pose parameters in geodetic coordinates by combining the pose angle and GPS information of cameras at the filming moment.At the end of this paper,we implement several intersection algorithms,and then evaluate and analyze the availability and accuracy of our algorithm on many groups of image sequences by shooting various kinds of targets in different places.