UAV Gesture Aerial Photography Control Based On Deep Learning

With the rapid development of UAV technology,UAV aerial photography has been widely used in various fields of life.Nowadays,the control of UAV aerial photography mostly relies on traditional remote control and hardware equipment,which has great application limitations,which has brought huge obstacles to the popularization and promotion of UAVs.In order to improve the maneuverability of close-range drone aerial photography,gesture recognition technology is introduced to make drone aerial photography more convenient and intuitive.This paper studies a gesture control system that can be applied to UAVs.This method has important practical significance in expanding the aerial photography of UAVs.First,the gesture recognition system is designed and implemented.The drone aerial camera shoots the current picture.In view of the influence of light during the shooting process,the method of light compensation is used to make up for the problem that the color detection rate of gestures drops in some special environments.Comparing the distribution characteristics of skin color in different color spaces,choose YCr Cb color space for skin color modeling.For the existence of noise and interference in the image,the methods of image binarization,image noise reduction,face detection and removal,and arm removal are used to effectively reduce other interferences in the gesture detection process.Aiming at the problem of slow gesture recognition speed and low accuracy,this paper compares different deep learning-based gesture recognition methods,and proposes Goog Le Net+PNN neural network to extract and classify the extracted gesture contour images,and obtain a good gesture recognition method.Accuracy and recognition speed.In the model training process,in view of the problem that the existing public datasets cannot meet the current model training problem,this paper collects and builds a dedicated dataset for model training and testing.Secondly,the drone gesture aerial photography control system is constructed.In view of the limited carrying capacity of the UAV,the microcomputer Raspberry Pi is selected as the onboard computer,and the Pixhawk is used as the flight control center.By building a communication module based on the MAVLink protocol,the direct communication between the UAV’s onboard computer and the Pixhawk flight control is realized,so that the on-board computer can read and send message frames effectively.By building a UAV vision module,the process of camera configuration,aerial image storage,image processing and gesture recognition is realized.By building a UAV control command sending module,the aerial photography flight of the UAV can be controlled according to gesture actions.Finally,the experimental environment of UAV gesture aerial photography control is built,including simulation environment and real environment.For the test of the communication module based on the MAVLink protocol,the test of the visual-based gesture recognition module and the overall function test of the drone gesture aerial photography,it is found through experiments that the functions of each module of the system and the overall gesture aerial photography control function meet the requirements.Finally,the performance of the system is tested,including the UAV gesture recognition time,the command response time,and the effective distance of UAV aerial photography.The experimental results show that the method in this paper can achieve close-range gesture aerial photography control.