Research On Pose Estimation Method Based On Machine Vision In Intelligent Warehouse Sorting

In recent years,due to the huge distribution demand caused by convenient and fast Internet shopping,the efficiency and cost issues in the warehousing and sorting process have received widespread attention from major enterprises.In the tradition al manual sorting operation,the sorter mainly distributes the goods according to the goods list and the goods distribution map,which has the problems of low efficiency and high cost.The machine vision-based pose estimation method can automatically achie ve the three-dimensional positioning of the target.Combined with the mechanical actuator,it can greatly improve the distribution efficiency and reduce the labor cost.This paper takes the automated sorting operation of goods in warehousing and distribution as the research object,and studies and designs a machine vision-based pose estimation method in the intelligent sorting system for the positioning needs of objects in three-dimensional space.The main research contents of this paper are as follows:1.Introduce the research background and significance of the subject in detail,and analyze the needs of automatic sorting in large warehouse management.This paper analyze the application status of machine vision servo technology in the industry,and summarizes the research status of domestic and foreign sorting systems and the development of core technologies.2.The design and working principle of the intelligent sorting platform are designed and analyzed for the storage sorting of the sorting platform,the r obotic arm and the end actuator,and the omnidirectional mobile chassis in response to the needs of automatic sorting.Strong mechanical arm and end actuator design scheme,and for the three-dimensional visual imaging scheme combined with the characteristi cs of the storage environment,the imaging and the installation scheme of the camera and actuator are selected.3.A pose estimation algorithm based on machine vision is studied.In response to the demand for actively acquiring the target space position,f irstly,the point pose model is used to analyze the pose expression form in 3D space,and the rigid body features of the target object are obtained by corner extraction.Then,by analyzing the camera imaging model,the mapping relationship between the corn er point and the predicted point is extracted to solve the pose.Among them,a prediction model based on a deep convolutional neural network is proposed for the prediction of the corner point.It is verified by feedback iteration and loss calculation on th e sample label The effectiveness of the network model in predicting projection corners is discussed.Afterwards,through the edge extraction and segmentation of the point cloud information of the prediction area to correct the grab points,the purpose of r ecognizing and extracting the position and pose of the target object from the image information of the integrated depth camera is achieved.4.To develop the pose estimation module in the automatic sorting system,the design architecture of the overall software system is first analyzed,and then for the low coupling and extensible requirements of the pose estimation module,the framework design,package packaging and management,file and database management are three Research was carried out in various aspe cts,and finally the package of pose estimation algorithm in the software system was realized and it has strong expansion characteristics.For the shortcomings of low efficiency and high cost in manual sorting,automated software was used to achieve the ta rget of the shelf.And posture information acquisition and transmission.