Pallet Recognition And Picking System Research Of Automated Forklift

With the rapid development of computer and robot technology,the widely application of forklift robots in the logistics industry marks that warehousing logistics is moving towards mechanization and intelligence.However,there is a large uncertainty in the position and posture of pallet stacking due to the influence of work flow,equipment precision,manual work,etc.,in unstructured warehouses.Forklift robots were required higher flexible picking ability for uncertainty target information.Three questions to be resolved: 1)Model of the problem of pallet picking and using state variables described position and orientation information of the pallet increases the degree of freedom and flexibility;2)Research on picking algorithm in unstructured environment,reducing algorithm constraints and increasing application scenarios;3)Research on the accuracy of pallet detection,analyzes and studies the factors affecting the accuracy of pallet pose estimation with the sensor characteristics.This article researched on warehouse forklift,built a warehouse self-moving forklift platform,established a warehouse picking model,and propose a pallet identification and picking system implementation method.The effective range and estimation accuracy of the pallet pose estimation method based on RGB-D sensor and lidar fusion are studied to realize pallet identification,improve the flexibility and applicability of warehouse automated forklift,and reduce the industrial application condition of intelligent logistics storage equipment.1)Design and construction of automated forklift platform.Through the simulator control and the original car controller connection,the forklift movement and attachment control.Design the hardware structure and layered framework of the forklift control system.The combination of hardware and software framework realizes the functions of pallet identification,picking,positioning,obstacle avoidance and automatic driving of the forklift truck.Established the foundation for picking up jobs.2)The warehouse picking model was established,and the expression model of the spatial positional relationship between each object and each sensor in the warehouse environment was derived.The problem of estimating the pose uncertainty of the pallet is converted into the estimation problem of the coordinate system parameters.The expression of the fixed position of the pallet was converted into a multi-degree of freedom representation of the pallet.A calibration method in warehouse environment was proposed to calibrate the transformation matrix between the coordinate systems.(3)Researched on pallet estimation algorithm based on single sensor.1)Pallet pose estimation algorithm based on the research of RGB-D sensor was studied and a classification template with the environmental characteristics of the pallet recognition algorithm was proposed.A mesh compression matching algorithm was proposed to reduce the interference of noise on matching and ensured the running speed of the algorithm.The algorithm can realize multi-pallet identification;when the detection distance is 1m~4m and the pallet angle is within ±25°,the detection accuracy of the pallet is ±101mm and the angle precision is ±6.07°.2)An algorithm was proposed for estimating the localization and angle of pallet with lake information,based on 2-D Lidar.A detection model was built to define the geometry relationship between detection region the number of lidar point(detecting the pallet foot),the width of pallet foot and the resolution of Lidar.Pallet Pose was got with improved incremental line extraction algorithm.he detection accuracy of the pallet is ±60mm and the angle precision is ±6°.(4)Research on Fusion Method of RGB-D and Lidar.1)Research on pallet pose estimation algorithm based on visual and laser fusion.Filtering laser data based on category and image recognition results.Through the pallet pose estimation method of the adaptive line extraction method,the picking system could estimate the pallet pose for long distance and large angle.The maximum lateral detection distance of the algorithm was 4000 m.The farthest detection distance was 7500 mm,the position estimation accuracy was ±180mm,and the degree estimation accuracy was ±6°.2)Research on the fusion strategy of pallet pose estimation methods based on RGB-D,lidar and visual laser fusion.Through the multi-threading algorithm and the back-end optimization thread,the estimation results were fused,with methods of the detection range filtering,the result filtering,and the spatial re-projection fusion.(5)The pallet picking comprehensive test of the forklift robot shows: The warehouse picking model and calibration method are applicable to the real warehouse scene.The pallet recognition and picking system had a pose estimation accuracy of ±60mm and an angle accuracy of ±3°.The pallet identification picking system can realize the workflow of pallet pose estimation-path planning-path tracking-picking,which meets the requirements of the warehouse operation.