Research On Visual Measurement And Spraying Quality Evaluation Algorithm Of Building Exterior Wall Surface

With the rapid development of science and technology,all kinds of exterior wall spraying robots are emerging.This emerging technology not only greatly reduces the cost of the industry,but also liberates manpower from hard,heavy and high-risk work.In order to realize the automation of spraying operation of exterior wall,a visual intelligent assistance guidance system is designed in this thesis to equip "eyes" for the building exterior wall spraying robot.Visual measurement technology was introduced to measure the window size of the target spray area.The measurement system combines optical testing technology,computer vision and deep learning based algorithms.Multi-sensors catch color and depth information of the area to be sprayed which are then fused for the next step.Combined with the full convolution neural network to extract the edge of the window in the image,and then obtain the window size to achieve non-contact measurement.Based on this,then guide the exterior wall spraying robot to work,and finally evaluate the spraying quality.The specific work content is as follows:(1)The measurement technology of monocular camera is described in detail and designed the measurement device.First of all,analyze the working environment of the system,and select the appropriate hardware device planar-array CCD cameras and phase laser rangefinder through comparison.Then the appropriate control chip is selected to design a multi-mode signal processing circuit.The circuit includes the main control chip module,conversion chip module,etc,and fuses data information obtained by two external devices.Analyze and design the layout and routing of the hardware circuit schematic and PCB drawings,and finally perform board making,welding,and performance testing.(2)The four transformations in camera imaging are reviewed,and the causes of camera distortion are analyzed.The CCD camera was calibrated by Zhang’s calibration method based on the Open CV camera calibration toolbox,and then obtains the distortion matrix.After distortion correction,the ratio K between the actual physical length and the pixel length is determined.This paper proposes a list method to find out the value of K based on the limitations of the working distance.Through experiments,the segment measurement is performed in the interval of 0.8m-1.2m.To get the K value.(3)Aiming at the fact that the texture and color of the external wall image in the actual scene are single,the feature extraction in this thesis focuses on the low-level information that can better describe the edges,and then proposes an improved U-Net network model to detect the windows in the image.The improved U-Net network is a shallow network structure that increases the number of skip connections,which can better increase the extraction of low-level information.The experiment proves that the measurement based on the improved U-Net detection combined with the K value can control the error within 1cm,and the overall accuracy of the system can meet the actual measurement requirements of the building exterior wall.Then,the super-pixel segmentation algorithm is used to evaluate the quality of the sprayed wall surface,and the unpainted uniform part is divided for reprocessing,thereby effectively ensuring the spraying quality.