Research On Recognition Method Of Working Environment And Pose Estimation Of Facade Cleaning Robot

Facade cleaning robots have a variety of structures,however,most of them are large,slow,and low efficiency.And there is not much improvement in intelligence.The main difficulty is the complex outdoor environment.Based on the images obtained by camera,this paper analyzes and recognizes the environment of the robot,and estimates the pose of the robot by line feature matching.During the cleaning operation,the cleaning robot needs to recognize the current environment.An image segmentation method is proposed to identify win dow frames and walls in non-cleaning areas.We analyze the characteristics of the curtain wall cleaning environment.Through the feature of the target in the image,the YUV color model is selected.By limiting the color channel threshold,the contour of the target area is obtained.The line edge is obtained by the LSD method and line merge algorithm.Furthermore,the window frame is obtained by the polygon creation and merging rule.As for the recognition of wall,the same method is adopted.To optimize the resource of the robot,a method of stains recognition on curtain wall is proposed.The recognition method of heavy stains and moderate stains is based on image sharpness.The fuzzy detection method of Harr wavelet transform is used to recognize heavy stains.The factors that have a greatest impact on moderate stains recognition are the uncertainty of background image and light intensity.The method of moderate stains recognition can be divided into three steps,the high-pass filter is used to enhance feature,an adaptive contrast enhancement method to obtain the background images,the spectral residual method is applied to recognize stains.In order to avoid obstacles and complete the cleaning work,for the robot,it is necessary to obtain the robot's posture.This paper proposes a robot pose estimation method based on linear feature matching.The linear feature descriptor LBD is established and the matching result is generated.Using the Pn L method combining points and lines to obtain the camera pose matrix.Through the position of the camera in the robot,the posture of the facade cleaning robot can also be estimated by the transformation matrix.By accumulating robot displacement,the position of the robot and the cleaning progress can be calculated.A cleaning robot and a lifting platform were built and the visual interactive interface was designed.The experiment verifies the effectiveness of target detection,stain recognition and pose estimation.The overall effect performed well.