Research On Visual System Of Modular Reconfigurable Cable-driven Parallel Robot Based On Deep Learning

Computer vision is widely used in robot technology.In this paper,some techniques in robot vision system are studied for the cable-driven parallel robot based on the theory of deep learning.The main contribution are as follows:(1)For image classification,a feature evaluation criterion for deep feedforward neural network is proposed.Taking the deep feedforward neural network as the feature extractor,a linear discriminant analysis of the extracted features is carried out,and an evaluation criterion is defined to quantify the learning ability of the deep network.The experimental results in CIFAR-10 dataset show that the feature evaluation criterion has a good evaluation ability for the distribution of features.(2)A fast localization algorithm based on computer vision is proposed to locate the moving components of circular or linear guideway robot.First,the target is designed,then the target is detected.Finally,the post-processing module is used to locate the target accurately.The experimental results show that the positioning error can be less than 1 °and the processing time of each frame is less than 30 milliseconds.(3)Aiming at the problem of limited resources of robot control system,the training model is pruned.And the model solidification and cross-platform call are realized.(4)Build the experimental system.On the basis of the existing robot experiment platform,a vision software framework based on TensorFlow is designed.And two groups of experiments are designed on the experiment platform to study the static error and dynamic tracking error of the positioning algorithm and analyze the source of the error.