Research On Humanoid Robotic Hand Operation Methods Based On Machine Vision And Machine Learning

Since the birth of the world's first industrial robot,the application of various robots has spread to all aspects of human society's production and life.And with the rapid development of artificial intelligence technology,people's requirements for intelligent robots have become higher and higher.In recent years,there have been more and more studies on the operation of humanoid robotic hands,and its main purpose is to provide intelligent service robots with more intelligence and more precise control for all sectors of society.To this end,this thesis studies the operation methods of humanoid robotic hands based on machine vision and machine learning.By combining machine vision,deep learning,humanoid robotic hands,etc.system modeling and simulation analysis and experimental exploration are carried out in order to greatly improve the intelligence and applicability of humanoid robotic hands.First of all,in order to realize the intelligent perception of the humanoid robotic hands to the working environment,this thesis uses deep learning methods to conduct experiments on the improvement of the object detection,recognition and positioning capabilities of the humanoid robotic hand system,and uses the convolutional neural network method Faster R-CNN(Faster Region-based Convolutional Neural Network)to conduct object detection,recognition and position training experiments.In addition,we also conduct experiments and improve the small object recognition success rate of the detection network.A feature fusion structure is proposed on the Regional Proposal Network(RPN)of the Faster R-CNN method,and the feature map proposal layers of different levels are connected to obtain more accurate small object recognition results.The loss function and related evaluation functions are improved to obtain more accurate recognition success rate.By testing the impact of different numbers of proposals on network performance,and using different datasets for training and testing,the object detection,recognition and positioning network used in this topic is obtained.A large number of experimental results show that the improved network has a certain improvement in speed and accuracy compared to the original Faster R-CNN network.For the structure of the humanoid robotic hands,the D-H(Denavit-Hartenberg Matrix)matrix description method is used to carry out modeling and kinematics analysis of the humanoid robotic hands,including forward kinematics analysis and inverse kinematics analysis,to realize the motion control and operation trajectory analysis of the humanoid robotic hands.The analysis of the working trajectory provides a theoretical basis for the movement of the humanoid robotic hands in the subsequent simulation experiments.This article also uses simulation software to build a pair of humanoid robotic hand and itsoperating platform and conduct a large number of simulation experiments.It systematically integrates the target environment image acquisition module,object detection and recognition and positioning module based on Faster R-CNN,and humanoid manipulator control model etc.At the same time,combined with the analysis of calibration experiments,the humanoid manipulator and its manipulator operating system and operating method proposed in this thesis are simulated and verified.A large number of simulation experiment results show that the entire humanoid manipulator and its manipulator working system have a good effect and have good application prospects.