Research On Intelligent Interaction Technology Of Hoisting Robot Based On Limb Action Recognition

Manufacturing industry is the cornerstone of modern industry.in recent years,with the development of information technology,intelligent technology and new energy fields,the continuous cross-integration of multi-disciplinary disciplines has further promoted the tremendous development of China's manufacturing industry.In 2015,China launched the "Made in China 2025" strategy for the implementation of the manufacturing power,emphasizing the acceleration of the development of a new generation of information technology and manufacturing technology,and promoting smart manufacturing.In the context of intelligent manufacturing and multidisciplinary integration,this paper combines robotics and computer vision technology to complete intelligent hoisting technology research.In large-scale lifting occasions such as ports,mines and docks,in the traditional case,the on-site lifting commander and the lifting driver are required to complete the lifting operation of the large cargo.The lifting driver completes the lifting by observing the limb command signal of the lifting commander.Operation,long-distance long-distance hoisting operation of the lift driver is prone to fatigue,affecting hoisting safety and lifting efficiency.In order to solve the above problems,this paper adopts the heterogeneous integrated learning method to complete the training and identification of the limb signal of the lifting command.The large zoom network camera automatically performs limb recognition on the lifting commander's lifting limb signal,and converts the identified lifting command limb signal into a lifting robot lifting operation through the robot control module.This method greatly releases the work of the lifting driver and greatly improves the lifting safety and lifting efficiency.Based on the marine logistics equipment intelligent loading and unloading platform jointly completed by Jilin University and Jimei University,this paper studies the method of limb signal recognition based on the heterogeneous integrated learning method.The main work of the thesis is as follows:1.The monitoring distance of Kinect is limited,and it can not meet the realistic complex environment and long-distance lifting application scenarios.This paper uses the large zoom network camera as the image acquisition device,and builds the OpenPose of Carnegie Mellon University human skeleton extraction deep learning framework to extract 18 human bodies.Skeleton node coordinate information.Data processing is performed on the extracted skeleton information to obtain preprocessed data,that is,skeleton vector and RGB skeleton map.In order to prevent the over-fitting phenomenon in the network training process,the training data set is expanded by using the rotation,translation,multi-scale scaling and affine in the RGB skeleton diagram;2.The hoisting command limb signal recognition result controls the movement of the KUKA robot and the hoisting mechanism by TCP/IP communication.Firstly,the forward kinematics analysis of the hoisting robot KUKA-KR16 is performed to coordinate the modeling,and then the inverse kinematics analysis is performed to calculate the rotational angle of each axis of the robot moving from the current position to the target position.For the hoisting mechanism,the characteristics of the three speed control methods: linear acceleration/deceleration method,exponential acceleration/deceleration method and S-curve acceleration/deceleration method are analyzed.Finally,the motion control of the hoisting hoisting mechanism is completed based on the five-stage S-curve acceleration and deceleration method.3.Constructing the hoisting instruction limb signal classifier,in order to exploit the BP neural network to extract shallow features and convolutional neural network to extract the abstract feature advantages,BP neural network and InceptionV3 network-based classifier are used to construct an integrated classifier based on the heterogeneous integrated learning method.Lifting instructions training and identification of limb signals.The BP neural network input is the skeleton vector,and the input data is initialized by the Xavier method.In order to better extract the skeleton vector point line feature,two hidden layers are set and the number of hidden layer nodes is updated by trial and error.Using the initial large learning rate and progressive reduction of learning rate to complete the training and testing of BP neural network;InceptionV3 network input is RGB skeleton vector,because the training of convolutional neural network requires a large data set,this paper uses limited The training data set uses the migration learning and Fine-tune methods to train and test the InceptionV3 network;4.Set up an experimental platform for hoisting robot lifting limb signal recognition system,and complete the hoisting robot controller,large zoom network camera and industrial control PC LAN.The analysis of the effect of skeleton extraction experiment and the experiment and analysis of the limb signal recognition of lifting command,then introduce the dual verification control method of the lifting robot.Finally,the software interface and function design are analyzed.