Design And Implementation Of A Gesture Recognition System For Wheeled Robot Control

With the development of human-computer interaction technology,gesture recognition technology is widely used in the field of robot control.In recent years,gesture recognition related algorithms have developed rapidly.Based on deep learning technology,this thesis studies the algorithm optimization,data set establishment,system implementation,etc.,and gives a gesture recognition method suitable for wheeled robot control.Based on this method,a wheeled robot based on gesture interaction is realized.The gesture recognition method in this thesis consists of a sequence structure of gesture detection,gesture key point location and gesture classification.In the aspect of gesture detection,the SSD network structure is improved by adjusting the network detection scale.In this thesis,the modified network structure is called SSD-GH.This structure reduces the number of detection frames in the SSD and improves the detection speed.In terms of point positioning,in order to speed up the positioning speed,this thesis reasonably cuts the convolution kernel size based on the CPM network architecture.In this thesis,the modified network structure is called CPM-GH;in the data set establishment,in order to better gesture In the classification research,this thesis builds two gesture classification data sets based on various backgrounds and illumination conditions,including 31234 gesture images,and trains the AlexNet gesture classifier based on this dataset to realize the classification and recognition function of 10 gestures;In terms of system implementation,this thesis implements a wheeled robot control system based on gesture interaction with ROS,and completes the basic control function of the wheeled robot.The gesture recognition method presented in this thesis is based on a monocular camera and has the advantage of low cost.Experiments show that the improved SSD-GH has faster gesture detection speed than SSD,the network forward time is shortened by 2.3ms,and the detection speed is increased by 11.2%;the improved CPM-GH is 4.5 times shorter than the CPM network forward time.Ms,key point positioning speed increased by 7.1%.At the same time,the human-computer interaction system of this thesis ensures the good real-time performance while completing the basic control of the wheeled robot,and the FPS reaches 11 or more.In terms of robustness,the average recognition accuracy of the system is over 99%.At the same time,the recognition range has reached 20cm-150 cm.In summary,this thesis designs a gesture recognition system suitable for wheeled robot control based on computer vision gesture interaction technology.It has the advantages of low cost,good robustness and good real-time performance,realizing real-time control of wheeled robots.And achieve a good balance in the pair of indicators of real-time and precision.After being extended,the system can be applied to various occasions,such as gesture control baggage handling,gesture control parking,etc.,and has certain practical application value.