Research On Online Identification Technology Of Production Behavior Based On Improved Skeleton Model

Recognizing and understanding human action is one of the main tasks of future workshop personnel safety management and control,and it is also the main difficulty and bottleneck of man-machine integration technology.There are many key production stations in complex workshops.Although the current workshop production monitoring system can be used for postevent accountability,it is difficult to achieve comprehensive,real-time,and synchronous control.In view of this,this paper proposes a method for identifying,analyzing,and managing the production action of key locations in the workshop based on computer vision technology.Based on the real-time action recognition project of production personnel at key locations in an aerospace workshop,this paper proposes an online production action recognition method based on the improved skeleton model of production personnel.First,the Kinect depth sensor collects different data streams such as color,depth,and human skeleton,and collects various data sets based on different tasks such as offline classification of workshop actions,online detection,and handheld objects.The skeleton data is improved through methods such as preprocessing and topological graph representation.Then,aiming at the problem of offline production action classification,a deep learning model based on graph convolutional neural network is constructed to realize intelligent recognition of production action.Aiming at the problem of online workshop action recognition,on the basis of graph neural network,combined with temporal convolutional network to construct online production action real-time detection model,and verify and analyze it on related dataset.In addition,considering that there is a certain interaction between the personnel in the production workshop and the environmental scene,based on the multi-stream data of the interaction of the workshop personnel,the area growth method and the neural network are used to identify the objects held by the production personnel.Finally,on the basis of the above algorithm,a real-time workshop production action recognition software and hardware system was developed to realize effective recognition of workshop personnel action.The above-mentioned methods have been experimentally verified in related data sets and project workshops,and have achieved good results.It has important research value and practical significance for standardizing the manufacturing process of the production workshop,reducing the hidden dangers of production safety,and realizing the multi-dimensional and multi-level monitoring,recording and management of manufacturing action.