Research On Intelligent Vehicle Control Algorithm Based On Gesture Recognition

Remote control intelligent car can replace human beings in the extreme environment for data collection,to improve work efficiency and personal safety are of great significance.The traditional remote control car is controlled by joystick or keyboard,which limits the flexibility of the car to a great extent.In order to solve this problem,this article will gesture recognition and the combination of remote control of intelligent car,through the camera to collect the user's gestures images and by gesture recognition algorithm to identify information,into the car control commands sent via wireless network to the car after car control unit,realize the intelligent car remote control based on gesture recognition.First of all,through the literature,the gesture recognition of the status quo at home and abroad were reviewed and summarized,on this basis proposes will gesture recognition with the combination of artificial neural network method,and further study of the yolo v1 based on neural network and yolo v2 target recognition algorithm,because yolo v2 in recognition speed and accuracy are better than the yolo v1,therefore,this topic selection yolo v2 target recognition algorithm to identify gestures.Second,in the use of network to do original image feature extraction of feature extraction,considering the complex network structure and the layer number is overmuch,easy to cause the error gradient disappeared in the gradient in the process of back propagation of this problem,proposed to improve the feature extraction of the network model,to overcome the error in back propagation problem of gradient disappeared,and reduce the network model of training time and improve the model of target positioning accuracy,makes up the defect of the traditional feature extraction network.At the same time,in order to determine the effective recognition distance of this algorithm for gesture,through a large number of data verification,combined with the 0.618 search algorithm,the final determination of the distance boundary of gesture recognition is 30～86cm,within this range,the mAP of gesture recognition can maintain above 65%.Thirdly,the control principle of the control end,namely the two-wheel self-balancing trolley,is studied carefully.In order to make the dynamic performance of the two-wheel self-balancing intelligent trolley more stable,the output of the gyroscope and the accelerometer is fused into the accurate inclination Angle and angular velocity output by kalman filter,which provides a strong guarantee for the control of the system.Experimental results show that the dynamic performance of the car has been greatly improved.Finally,the wireless communication network between the upper computer and the car is built,and the network delay and data transmission capability are tested.In order to verify the practical application ability of the remote control system,this paper connects the gesture recognition terminal with the smart car through the wireless network,realizes the real-time control of the smart car with gestures through the built user interface,and observes the surrounding environment information of the car through the monitoring interface.The design of the remote control system is reasonable,the control is flexible and the monitoring interface is clear.