Design And Research Of Visual Navigation System For Industrial Mobile Robot Based On Deep Learning

With the development of smart manufacturing and factories,the demands for industrial robot is constantly rising.Mobile robot is an important part of industrial robots.The application of mobile robot in production workshops has gradually become an important method to improve the economic efficiency of enterprises.The development of mobile robot has great value in improving efficiency,saving production time,saving energy and protecting the environment.The navigation system for mobile robot is one of the main direction and focus of current research.This research focused on image processing and intelligent algorithms.In this paper,the path tracking and positioning technology of mobile robot were studied.The main research contents are as follows:Firstly,the hardware components of the superior machine and the inferior machine,including the computer server,wireless server,vehicle controller,execute component,were designed based on the analysis on overall system architecture and working principle.And the software systems of server and client were designed and implemented.The overall system of mobile robot was constructed.Secondly,the image acquisition system was designed based on the feature of the experimental environment.Corrected grayscale images were obtained after pre-processing the acquired images.Then the lane line edge extraction and straight line fitting were performed to obtain the center lane line.The angle and distance of the offset between mobile robot and the center lane line were calculated in real time.Path tracking was implemented through adjusting the output signal to control the movement of mobile robot.Compared the sensor data with the results of path tracking algorithm,the deviation of the algorithm was within 1.5°.Finally,a QR code detection algorithm based on deep learning network was proposed.The deep learning detection network was trained through self-built training sets.For the twocategory problem of QR code,the deep learning algorithm proposed in this paper was compared with traditional machine learning SVM algorithm,the effectiveness and speed of the deep learning detection algorithm were verified.The average precision and accuracy of the deep learning algorithm reached 97.4% and 91.5% respectively.The average processing time per image was 41 ms,which could fulfil basic requirement of the real-time detection of QR codes for mobile robot vision navigation.