Research On Autonomous Flight Method Of Flight Control Platform For Road Traffic Flow Detection

With the continuous growth of car ownership,traffic congestion is increasingly becoming an urgent problem to be solved.And the real-time detection of traffic flow is the core to solve the problem of traffic congestion.By analyzing the advantages and disadvantages of the existing traffic flow detection methods,and considering the advantages of UAV such as strong mobility,low cost and good visibility,this paper proposes to use UAV to realize road traffic flow detection autonomously.However,the existing UAVs are difficult to fly autonomously in the weak GPS environment and do not have the function of traffic flow detection.Therefore,this paper studies the method of autonomous flight of flight control platform in the background of road traffic flow detection based on the automatic driving software PX4,aiming at realizing the autonomous flight of UAV and providing a solution for traffic flow detection.The following research has been done:(1)A real-time road target depth neural network detection model for flight control platform is established.Aiming at the existing target detection model can not meet the requirements of road target detection accuracy and real-time,and in order to improve the accuracy and real-time of road target detection,the real-time road target detection model is proposed based on residual block and multi-scale feature extraction,Logistic classifier to classify and Anchor to predict road target bounding box through analyzing the advantages and disadvantages of the existing target detection model.Finally,the accuracy and the real-time of this model is tested through experiments.(2)An unsupervised monocular depth estimation model for flight control platform is established.Inspired by the principle of binocular depth estimation,an unsupervised monocular depth estimation model is proposed based on reconstructing virtual camera to realize monocular depth estimation,which transforms the unsupervised monocular depth estimation problem into an image reconstruction problem.In order to improve the precision of monocular depth estimation,pyramid processing is carried out on the input picture,image reconstruction is carried out on a residual-based coarse feature extraction network and a deconvolution-based fine feature extraction network,and a joint loss function based on corresponding view reconstruction loss,parallax smoothness loss and depth map consistency loss is constructed.Finally,the effect of depth estimation of this model is fully verified through comparative experiments,ablation experiments and generalization experiments.(3)A deep neural network autonomous tracking model for flight control platform is established.A depth neural network autonomous tracking model for flight control platform is proposed through analyzing the advantages and disadvantages of the existing vision-based autonomous tracking model for quad-rotor unmanned aerial vehicles and the flight principle of quad-rotor unmanned aerial vehicles,which transforms the vision-based autonomous tracking problem for quad-rotor UAV into the problem of image classification.In order to improve the accuracy and real-time performance of the autonomous tracking model,the residual neural network is used for feature extraction,Leaky RELU is used as the activation function,and label smoothness loss is introduced into the cross entropy loss function.Finally,the accuracy and real-time performance of the model are tested through experiments.(4)The construction and experiment of the autonomous flight system of the flight control platform for road traffic flow detection are completed.In order to verify the feasibility of the autonomous flight system of the flight control platform for road traffic flow detection built by the above algorithm,considering the danger of directly testing the system on real roads,the method of combination of simulating experiments and real road experiments is proposed to test the performance of the system built in this paper.In order to facilitate the secondary development of quad-rotor UAV,the hardware platform is built based on Pixhawk flight control and NVIDIA Jetson TX2.In order to test the performance of the above algorithm,the software system is built based on ROS and PX4.Then the simulation environment is built based on Gazebo,and the reliability of the system is verified based on the simulation environment from PX4 firmware performance test,Offboard model reliability test,ROS upper layer algorithm test and video-based quad-rotor UAV simulation flight.Finally,experiments are carried out on real roads,and the experimental results show that the autonomous flight method of the flight control platform for road traffic flow detection proposed in this paper can realize the autonomous flight of quad-rotor UAV.