The Application Research Of Generalized Iterative Closet Point In Visual Odometry

The Visual Odometry is a system that uses the camera as a sensor input device and employs the estimation algorithm to realize the camera's self-motion estimation.It has broad application prospects in mobile robot,intelligent wearable device and augmented reality.At present,the main self-motion estimation method of Visual Odometry is feature point method.The feature point method has the disadvantage of difficult to extract image feature and declining pose estimation performance when the environment texture is not obvious.The release of RGB-D cameras is conducive to solving this problem.Through the point cloud data generated by RGB-D camera's depth image data,it can be used to estimate camera movement without relying on the environment texture.Point cloud registration,which is complementary to the motion estimation scheme of feature point method.This thesis uses Generalized Iterative Closet Point to realize point cloud registration for conditions with unconspicuous texture.In this way,the motion estimation of the Visual Odometry is carried out in this scene,and the adaptability of the Visual Odometry to the untextural scene is enhanced by using this method.Applying the point cloud registration method to estimate the motion of the camera requires point cloud data of the environment.In RGB-D camera,the internal parameters of the camera model are used to combine the depth images to generate the point cloud data.The calibration algorithm is used to calculate this parameter to obtain the internal parameters of the camera model.This paper analyzes the RGB-D camera model in detail,and a convenient camera calibration program is designed on this basis.The program can shoot and display the color image,depth image and infrared image data collected by RGB-D sensor in real time.On the basis of obtaining the calibration plate image,it is possible to adjust the position of the detected corner by adjusting the slider on the program interface to facilitate the process of calibration.Secondly,on the basis of obtaining the camera model parameters,this paper analyzes the algorithm principle of the Generalized Iterative Closet Point in detail.Then,for the optimization objective function generated by matching pair of the nearest points,this thesis apply Lie algebras and combine the nonlinear least square method to solve the problem.Then,according to the optimization objective function constructed by matched closest points,lie algebras parameter pose is applied,combining the nonlinear least square method,to solve the problem.The performance of this algorithm and the original algorithm are carried out contrastive analysis.On the basis of applying this algorithm,combined point cloud down sampling,Generalized Iterative Closet Point is used to deal with the motion estimationproblem of the Visual Odometry in the unconspicuous texture.The odometry system was tested on the open data set.Finally,the actual scene experiment of the Visual Odometry applying Generalized Iterative Closet Point is carried out on the actual Kinect V1 sensor.The correctness and stability of the system is verified,and the motion trajectory obtained by estimates is analyzed.