Research On Operation Area Perception Algorithm Of Overhead Crane Based On Vision

Area awareness for overhead cranes ensures safe and efficient operation of overhead cranes.By sensing where an overhead crane is operating,operators can avoid collisions,accidents,and other errors that can reduce injury,equipment damage,or costly downtime.They can also optimize the movement of the crane and its load,increasing productivity and reducing the risk of errors or inefficiencies.With the adjustment of the national strategy and the increasingly intensified market competition,the trend of overhead cranes towards automation,intelligence and unmanned development is becoming more and more obvious.There are some special difficulties in the visual perception of the overhead crane operating area.Specifically,the operating area of general bridge cranes is outdoors,which is easily affected by factors such as light intensity and weather changes.On the other hand,target tracking and detection is timeconsuming in the visual perception system,so it is necessary to design a lightweight algorithm that can run quickly and efficiently to ensure the real-time and robustness of the perception system.In this paper,the research on the visual perception algorithm of the bridge crane operating area has been done as follows:(1)This paper studies the overhead traveling crane working area systematically,summarizes the difficulties and challenges of overhead traveling crane working area vision perception,and puts forward the requirements for the algorithm design in this paper.Summarize the basic knowledge of deep learning and convolutional neural network,summarize and analyze the classic feature extraction network,typical target detection method,typical semantic segmentation method,and provide the basis for the algorithm design of this paper.(2)Furthermore,a lightweight feature extraction network based on adaptive attention mechanism is designed.In this paper,based on the research of multitasks based model,we propose a medium multi-scale multi-task visual perception method.The main tasks of this network include: object detection,semantic segmentation,motion segmentation,depth estimation and visual ranging.To address the phenomenon that jointly trained models perform better than models trained with a single task,this paper uses multiple tasks trained in succession,with each task supporting each other.The study on the detection of moving objects in the operation area of bridge cranes reveals that the scale of moving objects varies greatly.To address this problem,a shared encoder with multi-scale feature fusion is designed in this paper,which both improves the network performance and reduces the computational effort.(3)Based on the research of multi-task model,a multi-scale multi-task visual perception method is proposed in this paper.The main tasks of the network include: object detection,semantic segmentation,and motion segmentation.In order to solve the problem that the performance of the model trained jointly is better than that of the model trained by a single task,this paper uses multi-task training to support each other.The research on the detection of moving objects in the working area of overhead traveling crane shows that the scale of moving objects varies greatly.To solve this problem,a multi-scale feature fusion shared encoder is designed,which not only improves the network performance but also reduces the computational complexity.(4)An experimental platform based on the TDA4 edge computing module is built,and the algorithms designed in this paper are deployed into it to verify the feasibility of the algorithms.The algorithm deployment scheme was designed.The final deployed model inference speed and detection accuracy are 232 FPS and 80.1%.The visual perception algorithm designed in this paper can well perceive and detect objects in the operation area of overhead cranes,laying the foundation for the intelligent and unmanned overhead crane system.