Research And Implementation Of Navigation And Pallet Positioning Method Of Unmanned Forklift Based On Vision

In Intelligent Logistics System,automated equipment such as unmanned forklifts can solve the problems of serious aging and labor shortages,and at the same time,it can improve efficiency in warehousing,handling and other links and reduce the error rate of these links,so it plays a very important role in intelligent warehousing.This thesis takes forklifts as the research object to study the navigation of unmanned forklifts in natural storage environments and the precise positioning of pallets.The lane is a natural part of the natural storage environment,and the lane detection is also one of the most important parts in automated driving,so the lane is used as the navigation basis.Because the lane has a unique structure and is easily affected by a variety of complex environments(such as light,occlusion,blur,etc.),lane detection is also a very challenging task;what is the difficulty of pallet positioning on the premise of ensuring real-time detection,the detection accuracy is improved,and it is used for actual forklifts loading and unloading of pallets.Regarding the above difficulties,the main content and research results of this thesis are as follows:1.Aiming at the accuracy and real-time problem of lane detection under extreme conditions,this thesis proposes a lane detection algorithm based on spatial feature aggregation.It is difficult for traditional convolutional neural networks to directly learn fine lane spatial features.This thesis uses the spatial feature aggregation module to fuse and enhance the features extracted by the convolutional neural network in spatial dimensions,providing rich spatial features information for the cascaded decoder,and proposes an up-sampling decoder that combines bilinear interpolation and deconvolution to restore the feature map to the original image size,and perform pixel-level segmentation on lane.The experimental results prove that the spatial feature aggregation module obtains fine global information by aggregating the horizontal and vertical feature maps,and the up-sampling decoder effectively restores the lane features,which can improve the performance of the lane detection algorithm in a variety of complex environments,and will not affect the speed of detection.2.At present,the main pallet positioning uses radio frequency tags or QR codes,etc.,which are complicated and troublesome and costly.This thesis proposes a pallet positioning method that combines object detection and image analysis.Based on the object detection results,image analysis is used to further accurately locate the pallet in the image.The experimental results prove the effectiveness of the method and meet the actual engineering requirements.3.In order to verify the applicability of navigation based on lane detection and pallet positioning,this thesis designs a simple experimental site and uses an unmanned forklift as the research object.The forklift uses the lane detection result as the navigation basis,and executes the loading and unloading tasks of the pallet according to the pallet positioning result to realize the point-to-point pallet handling.The experiment proves the feasibility of the scheme.