Real Time Recognition System Of Stacking Information Based On Machine Vision

As electronic trading technology continue developing,the demand for intelligent storage is expanding day by day.At present,the traditional counting method to realize intelligent storage is mainly mechanical counting and RFID counting.Traditional smart counting methods have various limitations such as high cost and lack of accuracy.In the era of rapid development of artificial intelligence,machine vision technology has played an irreplaceable role in the development of all walks of life.Due to the superiority of machine vision methods,it has broken many technical barriers that cannot be solved by traditional methods.In view of the above reasons,this system has developed a real-time recognition system of stacking information based on machine vision to accomplish the goal of storage counting through the recognition of cargo stacking information.Therefore,this thesis mainly focuses on an in-depth study of a real-time stacking information recognition system.Before the system runs,an image acquisition frame is built to obtain the image information of stacking.After the system is started,the target detection function is carried out for the forklift truck loaded with stacking cargo,and the image of the optimal position is selected and entered into the image segmentation part.The system divides the image into corresponding regions,and then calculates the size and quantity information of the cargo box according to the calculation method designed.During the operation of the system,the forklift only needs to transport goods past the image acquisition frame and does not need to stop specifically to complete subsequent operations,so the system can achieve the real-time requirements.In addition,a 20 degree tilt of the forklift is allowed to pass the frame,which ensures the robustness of the system to some extent.At the same time,this system provides the function of in/out warehouse status discriminating and information recording with clear interface,which can complete the process of real-time identification of stacking information as well as storage recording.Considering that the specifications of cargo stacking may be various and occasionally there are forms of different sizes and multi-layer stacks,but due to the current technical problems,it is impossible to accurately identify and detect the complex internal structure of stacks.Therefore,after investigating the actual operation of the factory,this thesis only targets the information of single-layer stacking with the same specifications,while facing the rare occurrence of stacking with different specifications or multi-layer stacking,it is necessary to make partial adjustment manually before identifying and counting them.After years of development of machine vision technology,various types of deep learning networks have been able to complete the function of target detection and image segmentation well,and how to choose the network to take into account real-time,robustness at the same time,with good performance has become the core idea of this thesis research.The system does not require high hardware equipment.It only needs a frame equipped with two depth cameras to complete the construction of image acquisition.It is simple and easy to operate,which greatly solves the problem of high labor cost.And this has great practical value for the factory storage application.