Research On Visual Recognition Technology Of Logistics Pallet In Complex Environment

Logistics pallet intelligent identification technology is of great significance to improve factory automation level and production efficiency.However,existing technologies are mostly limited by the complex environment of factories,or cannot meet the requirements of high accuracy,real-time and stability of pallet detection and positioning in production.Therefore,designing an algorithm with high overall detection accuracy and high speed has become an urgent problem to be solved.In this paper,according to the actual factory environment of Zhejiang Zhongli Machinery Equipment Co.,LTD.,using target detection and location technology,a high precision,fast speed,high stability of logistics tray detection and location algorithm is proposed.The main contents of this paper include the following aspects:Firstly,the article analyze the automatic forklift working conditions and the actual demands,design algorithm and experiment scheme,collect pallet data sets and the depth of the corresponding image in multiple factories or production workshop environment,analyze image features,and annotation data set.Secondly,logistics pallet rapid detection,because this article get pallet data sets collected image environment background is extremely complex,dynamic and static obstacles,final positioning of logistics tray will exist great interference,so at first,this paper design a fast detection algorithm of pallet,by improving YOLOv5 s network carries on the test on the pallet in the image positioning,and pass the lightweight network model of over-pruning and other methods can improve the accuracy of the algorithm by adding small target detection layer and finally complete the detection and positioning of the pallet on the original image.Thirdly,the rapid detection of the pallet jack in the pallet image,the pallet jack is the most special features of the pallet,and also is the most need to focus on regional automatic forklift,therefore very suitable for testing the special point on the pallet to extract the image pixel coordinates,this paper adopts SOLOv2 ultra-light network,detection of pallet jack in pallet images,put forward four models,can adapt to a number of different requirements of the environment,and finally complete the detection of pallet jack,extraction of the pallet jack endpoint pixel coordinates on the pallet image.Forth,logistics pallet positioning module,the module will be above the pallet jack to extract the pallet jack endpoint detection module pixel coordinates on the pallet icon,first convert to the original image pixel coordinates,then extract the depth of the original image corresponding to the image,through the depth of the image pixel coordinates extracting depth information,and then through the camera parameters and the coordinate transformation formula,the pallet jack endpoint pixel coordinate transformation on the pixel coordinate system to the camera coordinate system under the pallet jack endpoint of three-dimensional coordinates,and then through the pallet jack endpoint and target distance information,or in accordance with the pallet fixed size information to calculate the target of three-dimensional coordinates,as well as the camera and the calculation of the pallet jack section plane angle,get a pallet location information and position information.Finally,through experiments,the proposed logistics pallet detection and positioning algorithm has high detection speed and not only has high image pixel accuracy,but also has millimeter-level positioning accuracy.In the pallet detection stage,the accuracy is up to 99.7%,the time is only 5.2ms,the speed is up to 192.3 FPS,and in the pallet jack detection stage,the accuracy is up to 92.5 %,the lowest is 91.8%,the fastest time is only 27.6ms,the slowest is only 41.4ms,the overall detection accuracy of the algorithm is 91.7%,the time is only 41.3ms,the maximum average absolute value error of positioning accuracy is only 9.48 mm,to achieve efficient and high precision positioning of logistics pallet.