Research Of Machine Vision And Its Application On Automatic Fueling System

Petroleum plays a very important role in the national economy,it is known as black gold and the blood of industry.Usually,petroleum is refined to obtain other chemical products,which is stored by the place of origin and then shipped to the destination by the tank trucks,such as lubricating oil.The height of the tank truck is 2.3?4.2m and the position of the jar is on the top of the tank body.Because of the limitation of the height of tank truck and the position of the jar,there are two ways to fuel,the first way requires workers climb to the top of the tank truck,and then push the pipeline above the tank truck jar by hands;the second way requires the driver to stop the tank truck directly below the oil pipeline.There are potential safety hazards during the worktime of workers,and accidents may happen due to the negligence of workers.Based on machine vision technology and machine learning algorithms,this paper researches on the key technology of positioning of jars automatically,then designs a machine vision system capable of positioning accurately.The main contents of this paper includes the following aspects:Firstly,using traditional machine vision image processing method to find the center of the tank jar with the pictures collected by the camera.First of all,converting the RGB images into gray-scale images,because the single channel images are convenient for subsequent calculations and improves the speed of calculations.Then,by comparing different image filter methods,using bilateral filter to eliminate image noises,and then applying global threshold segmentation to extract the candidate area of objects.Next step is using close morphological transformation on the target areas to remove some interference areas,and then selecting the largest area is considered as the area of object by sorting areas.Finally,the pixel coordinates of the center of the jar are obtained by the minimum enclosing curve fitting method,and then calibrating by the nine-point calibration method,which could transfer the pixel coordinates into the mechanical device,and the machine moves to the specified position to complete the fueling action by the WMX motion control system.Secondly,because of the influence of environmental factors in traditional machine vision,it is proposed to solve the problem by using the combination of jar contour extraction based on Canny operator and the unsupervised machine learning algorithm K-means.First of all,Canny operator,comparing different image edge extraction methods,is used to extract edge contours for obtaining clearer and more continuous contours.Using K-means algorithm to cluster all the pixels,and set 5 clusters to obtain image clustering map.By combining edge features and clustering features,and then selecting the longest contour among combined areas,the target contour can be obtained.Finally,in order to improve the robustness of the algorithm and further eliminate the influence of illumination and oil stain,the target detection method based on HOG features and machine learning algorithms is proposed.First of all,the preprocess of the collected pictures includes obtaining positive and negative datasets,classification and data augmentation.Then,HOG features is chose which is used to reflect the image local texture by comparing different image features.Comparing the accuracy and IoU of SVM and FLGBDT algorithm in object detection,the FLGBDT algorithm has higher accuracy and finally is used as the classification model.According to the final bounding box obtained by the classifier,traditional machine vision method is applied to detect the objects.The research on the combination of machine vision and machine learning could solve the problems in the fueling process of tank trucks,which has theoretical value and practical engineering significance.