Research On Intelligent Detection And Control Technology Of Excavator Bucket Based On Vision

Excavator is a kind of construction machinery which is widely used and complex in structure.At present,many research institutions at home and abroad are committed to improving the automation and intelligent level of excavators.It is regarded as one of the important directions of the development of excavator to realize the autonomous digging function,which also promotes the development of the robot technology of excavator.Improving work efficiency,reducing labor intensity and adapting to complex environment have become the main research objectives of independent mining.In this thesis,the intelligent visual detection and control technology of mining excavator bucket is studied,so that the excavator can have the ability of environment perception in the actual working condition and reduce human intervention.Based on the location information and mining target area information of the bucket,the bucket mining track is planned without considering the rotation.Relying on vision-related technology to solve problems such as bucket identification and detection,mining target area detection and so on,the bucket trajectory planning and control technology after providing visual information is studied.The main research contents are as follows:According to the working conditions of mining excavator,a visual perception system based on the integration of lidar and monocular camera is designed and built,the intelligent detection function requirements of excavator bucket are analyzed,the visual sensor selection is made,and the installation bracket of the visual sensor is designed.Based on the industrial camera imaging model,monocular camera calibration is carried out to solve the distortion coefficient of internal and external participation.The joint calibration model of lidar and monocular camera is analyzed,and the relevant parameters of joint calibration are solved by Autoware to realize the spatio-temporal synchronization of lidar and monocular camera.Study the registration model of point cloud and image,realize the mapping of three-dimensional point cloud to twodimensional image,and provide feasibility for subsequent two-dimensional detection and three-dimensional detection association.The framework of bucket image detection is built based on Yolov5 s algorithm,and the bucket data set is constructed for training.The bucket image detection model is determined according to the bucket detection results and performance indicators,and the two-dimensional image detection frame and image features are output.According to the actual working conditions of the excavator,after ground filtering,downsampling and preprocessing range division of the point cloud collected by the lidar,the characteristic point cloud of the excavator bucket is extracted through the point cloud clustering,and the three-dimensional surrounding box and real-time position of the bucket are output.A fusion vision detection scheme is proposed which can accurately identify the spatial information of excavator bucket.Combined with the kinematic modeling of the excavator working device,the forward kinematics and reverse kinematics analysis of the mining excavator are completed,the conversion relationship between the joint space and the drive space is solved,the target mining area is detected,the initial position and target end point of the bucket are used to plan the bucket mining trajectory,and the bucket reached the target mining area by track tracking..The experimental results show that the intelligent detection scheme of excavator bucket based on vision has high accuracy and stability.It can use visual information to plan the mining track of the bucket without considering the rotation and realize the track tracking control,which can provide technical support for the intelligent operation of excavator.This thesis has 82 figures,22 tables and 92 references.