A Vision-based Localization System Applied To Concrete Floor Leveling Machine

With the rapid development of the construction industry,the current construction equipment is difficult to meet the actual needs of the project.It is the general trend to develop high-quality,high efficiency,miniaturization and automation construction equipment.The construction of concrete floor is an important work in the construction.At present,the concrete floor levelers used in the construction are all operated by workers,which costs a lot of manpower,and the personal safety of workers is difficult to guarantee.In view of the above problems,this paper designs a set of vision positioning system,which arranges multiple cooperative targets at the edge of the construction site,obtains the feature point information on the target through the camera to take the target image,and calculates the position and attitude between the camera and the target,so as to realize the self positioning of the leveling machine in the site.The vision-based localization system designed in this paper was mainly composed of camera,cooperative target,IMU and corresponding programs.The second chapter mainly introduced the imaging model of the camera and the design of the cooperative target.The cooperative target had been designed as a plane target,which was composed of an external rectangular contour and internal circular areas.The third chapter mainly introduced the target image recognition algorithm.It could be identified by median filter,gamma transform,threshold processing,Canny edge detection,and feature points could be extracted by the center of gravity method.Experiments showed that when the camera was 8m away from the target,it could effectively identify the target and extracted feature points.The fourth chapter mainly introduced the PNP algorithm.The internal parameters of the camera were obtained,and the imaging model of the camera was established.The pose of camera was solved by PNP algorithm,and the minimum binary problem was constructed by nonlinear optimization.The pose of camera was optimized to minimize the reprojection error.The fifth chapter mainly introduced the multi-sensor data fusion algorithm and positioning accuracy test.The Kalman filter and Unscented Kalman filter were used to optimize the observation data of the camera and IMU respectively,and the data fusion was carried out to obtain the camera’s pose in the actual work process.The positioning accuracy test showed that the maximum positioning errors in X and Y directions could be 0.2m and 0.5m within 5m distance between the camera and the target.When the distance between the camera and the target was 8m,the positioning errors in X and Y directions could be 0.5m and 1m.