Crane Climbing Robot Lane Line Recognition And Deviation Warning

In China,there are a large number of large cranes entering the late service.If cranes are manually inspected and analyzed and maintained,not only are they inefficient and costly,but they are also manually unable to carry out safe and effective work at high dead ends of cranes.Crane climbing robots can carry a variety of detection equipment and go to places that are inaccessible to human hands by attaching to the surface of the metal structure of the crane.To develop the automatic path finding function of the climbing robot,the implementation of lane edge line recognition and lane deviation warning are two crucial prerequisites.The main research of the thesis is as follows.(1)A lane line detection algorithm based on an improved image processing method is proposed for the recognition of lane edge lines of the climbing robot.The acquired image data is divided into four categories of scenes based on different features of interference information in the lane scenes.In the image preprocessing,a variety of grayscale algorithms were compared,and the experiment proved that the weighted average grayscale algorithm proposed in this thesis for crane color improvement works best.Finally,a lane line constraint algorithm based on the improved working condition of the climbing robot is proposed,and the straight lines obtained by Hoff detection are screened by multi-conditional constraints using this algorithm.The experimental results show that the lane line detection algorithm based on the improved image processing method has a higher recognition success rate than the other two traditional image processing methods in identifying the lane edge lines of the climbing robot.(2)An innovative improved U-net-based lane line recognition method is proposed.This thesis combines the features of U-net and Mobilenet,and proposes an improved network model M2-Unet.The M2-Unet network can achieve light weight while ensuring a good fit to a small sample of climbing robot lane data set.The test results on the test set show that segmentation of lane images using M2-Unet has high accuracy.Then the binarized lane images obtained by segmentation are extracted using the LSD straight line detection algorithm to extract the edge straight line segments and fitted using the least squares method,and finally the lane edge line extensions are excised and screened to remove the interfering straight lines to obtain the detection results of the left and right lane edge lines.By conducting recognition experiments on a large amount of image data,the quantitative evaluation of the experimental results shows that the algorithm proposed in this thesis can meet the accuracy requirements of practical applications.(3)A lane departure warning model for climbing robots based on dual safety thresholds is proposed based on the characteristic attributes of crane metal structure and climbing robot working conditions.This model is divided into two parts of safety threshold constraints,the first part takes the slope of the left and right lane edge lines as the constraint,and the second part takes the distance between the center point of the body of the climbing robot and the left and right lane edge lines as the constraint.By combining the two parts of judgment conditions,the early warning can be issued for the deviation of the body and the front of the climbing robot,respectively.