Real-Time Human-Posture Recognition For Human-drone Interaction Using Monocular Vision

Today,the drone is widely used in a variety of industries and plays an important role in many fields,such as military,agriculture,service industry,mining and so on.To meet the needs of different fields,functions and performance of the drone are different,and the drone system is also quite different.The interactive system between humans and drones is an important part of the drone system,which determines the operation experience of drone,and has high application value and research value.At present,most of the drone systems realize the interaction between humans and drones through the ground station.The human command is transmitted through the ground control station,remote control,mobile phone software,etc.The advantage of this method is that it has a wealth of technical accumulation and can achieve precise control.The disadvantage is that it relies on interactive devices and is not convenient enough.There are also some drone systems try to control the drone using human gestures.this way uses the inherent characteristics of human beings,it is more in line with human interaction habits and can achieve a better interactive experience.The disadvantage of this approach is that the effective distance of gesture recognition is too short.Therefore,there is a need for an interactive method that is more convenient and has a longer effective distance.Based on these problems,the paper proposes to use human posture to realize the interaction between humans and drones.The paper designs a human posture recognition algorithm for drones.The algorithm extracts the coordinates of the human joint in the image through the neural network and then calculates the distance between the joints to obtain the feature vector.The feature vector is normalized by linear calculation,and finally,the human posture is recognized by the support vector machine.Experiments show that the average accuracy of the algorithm is 97.34%,and the human posture can be effectively recognized at a long distance.In the actual environment,multiple human bodies may appear in the picture,causing serious interference and a large number of meaningless calculations.To solve this problem,the paper proposes to use the object tracking algorithm to track the drone controller in the image.This paper evaluates the existing object tracking algorithm and tests the success rate,accuracy and real-time of the object tracking algorithm on the data set.Finally,the KCF algorithm is selected for object tracking.Based on the above two works,a vision-based human posture instruction recognition system for the drone is designed and implemented in this paper.The image data collected by the drone is processed to identify the human body posture in the image,and then the human body posture is parsed into corresponding instructions;the target tracking algorithm is used to track the controller in the image to avoid interference from unrelated persons.In this paper,human posture is used to control the drone to achieve a more natural way of human-computer interaction.