Research On Key Technology Of Cargo Stacking Status Detection System Based On Machine Vision

During the transportation process of goods in a warehouse,improper manual handling and mechanical damage may result in damage or loss of goods.This not only affects the subsequent transportation and sale of goods but also leads to an increase in operational costs for businesses.Therefore,goods inspection has gradually become an important business in industries such as logistics and manufacturing.Traditional manual inspection methods have limitations such as limited detection range and insufficient accuracy,while machine vision technology has many advantages such as speed,accuracy,and non-contact and has therefore been widely used in various fields,gradually replacing traditional manual inspection methods.This thesis takes the cigarette boxes in the temporary transfer warehouse of a cigarette factory as the research object,and conducts relevant research on the problems of missing pile quantity and abnormal pile attitude caused by frequent transfers of goods in the warehouse.A products pile condition detection system based on machine vision is developed to meet the warehouse’s needs for goods detection.This thesis mainly studies the system’s software and hardware design,pre-processing of goods images and point cloud data,feature contour detection,template matching,and point cloud registration.A multi-scale feature template matching method is adopted to achieve missing goods quantity detection,and a point cloud registration algorithm combining fast point feature histograms（FPFH）,super four point fast robust matching algorithm（Super-4PCS）,and iterative closest point（ICP）were proposed to achieve goods pile attitude estimation.The specific research contents are as follows:（1）Data preprocessing methods are studied,including cargo color image preprocessing and 3D point cloud data preprocessing.In order to reduce the impact caused by insufficient and uneven lighting in the warehouse,the image of the goods is preprocessed,and image processing techniques such as bilateral filtering and histogram equalization are used to improve image quality and contrast.Experimental results show that the image preprocessing method can effectively balance the image chromatic aberration caused by uneven illumination and improve image contrast.Straight-through filtering and statistical filtering are used in the preparation of 3D point cloud data to remove noise points that were present in the obtained 3D point cloud.The experimental findings demonstrate that this preprocessing technique is capable of successfully removing both discrete and background point clouds from the data.（2）In this thesis,we study a method to detect missing cargo quantity based on multi-scale feature template matching.An improved Canny operator was employed to extract the feature contour of the brand on the surface of cigarette boxes,which was then used as the template image.The template image was used to construct a multi-scale image pyramid and then rotated.Finally,template matching was used to count the matched regions in the target area of the goods image.Experimental results show that the algorithm achieved a detection accuracy of 94.8%and an average detection time of 1.5seconds,which met the demand for detecting missing items in warehouses.（3）This thesis presents a novel method for estimating the stacking posture of goods based on point cloud registration.To achieve a robust point cloud coarse registration result when the point cloud data have incomplete overlap,a Super-4PCS+ICP point cloud registration method incorporating FPFH was proposed.This method reduces the point cloud volume by extracting FPFH feature histograms,and employs the Super-4PCS registration algorithm for coarse registration to obtain an initial alignment,followed by the ICP algorithm for fine pose adjustment during the fine registration stage.The experimental results demonstrate that this method can ensure registration accuracy in the case of incomplete overlap,with a root mean square error（RMSE）accuracy of the order of 10^-4.The method can effectively achieve point cloud registration and posture estimation for a single stack of cigarette cartons.A software system for detecting the state of goods stacking was designed and developed based on the Halcon visual processing library and the point cloud library（PCL）.The system was developed using the C#language and constructed on a Win Form framework.Experimental testing of the system show that it achieved an accuracy rate of over 90%for detecting missing goods,meeting the demand for missing goods detection.With respect to estimating the posture of goods,this system is capable of accurately registering the point cloud of goods,enabling effective estimation of the stacking posture of the cargo.It meets the requirements for estimating the posture of cargo stacking.