Research On Improved Gesture Recognition Algorithm And Its Application For Human-robot Collaboration

With the rapid development of artificial intelligence,big data,computer vision and other technologies,computer vision and human-computer collaboration technology has been more and more widely used in the manufacturing industry,enterprises also put forward higher requirements for human-computer collaboration in the intelligent manufacturing environment.Relevant technology mainly on man-machine coordinated through between man and computer access to information,identify,and conversion and application of all kinds of devices for operation control,the gesture since it has simple operation,convenient and flexible features similar to that of the normal operating mode,gradually became one of the key areas of humanmachine collaborative research,got the attention of more and more widely.The traditional gesture recognition method based on image preprocessing combined with machine learning is vulnerable to the influence of complex background or illumination conditions,which affects its accuracy and robustness.At the same time,the recognition distance cannot meet the requirements of environmental protection equipment enterprises due to various conditions.However,some gesture recognition methods with the help of external equipment such as data gloves have high accuracy,but the cost of equipment is high and the use is not flexible enough,which often restricts their application deployment in enterprises.In view of the above deficiencies,based on the convolutional neural network,this paper proposes a gesture recognition algorithm based on the improved SSD(Single Shot Multi Box Detector and OpenPose technology,and applies it to the simulation control of industrial robots in environmental equipment enterprises to achieve human-computer collaborative application.The main work contents of this paper are as follows:First will be a prerequisite for the SSD feature extraction network VGG-16 can be replaced by using depth volume classification Mobilenet product,improve the detection speed,simplify the recognition scope,the concentrated to hand area,through the K-means cluster analysis to predict box aspect ratio,optimize the ratio of high to width of predict box,and moderately expand the predicted range,makes the improvement of the SSD can better identify hand position;The method of data enhancement was used to expand the data set,and the transfer learning method was used to reduce the training time of improved SSD.Then,the improved SSD algorithm was set as the front network to extract hand images,and the hand images were input into the OpenPose hand key point model to construct the information of hand key points.Gesture classification algorithm was designed according to the spatial position of key points in the above key information to classify and recognize gestures.Eight kinds of conventional gestures were designed for gesture recognition.Based on the experimental verification method,the above gestures were compared.Compared with the unimproved SSD algorithm,the detection speed and training time of the improved SSD algorithm were greatly reduced.Compared with the Openpose before improvement,the experimental results show that this method can greatly improve the recognition distance of the model in a complex background,and can effectively enhance the robustness and accuracy of the model.Moreover,the application conditions of this algorithm are more flexible,which can improve the requirements of man-machine collaboration in long-distance conditions.Secondly,a human-machine collaborative virtual simulation environment was constructed,and the algorithm was applied to the industrial robot human-machine cooperative operation application in the virtual environment,and the application scene was verified.The framework of the application system includes three sub-function modules: data acquisition and analysis system,man-machine cooperation system and robot action system.Gesture recognition part as a data acquisition and analysis system.In this paper,five gestures with better effects were selected from the above eight gestures to design the corresponding relationship between gestures and industrial robot actions.The robot controller and industrial robot production environment were simulated through WEBOTS,and the robot action system was used to control the grasping,moving and other operations of the industrial robot.Finally,the simulation verification was carried out in the virtual environment of remote control and cooperation in a laboratory with a complex background environment similar to the manufacturing environment of environmental protection equipment enterprises.The verification results show that the control system can accurately identify gesture actions and accurately control industrial robots for cooperative control,achieving the expected goal of the paper.