Research On The System For Vehicle Detection And Tracking Based On Aerial Photography Used Rotor UAV

In recent years,vehicle detection and tracking technology has developed rapidly in domestic and abroad.It has been well-applied and popularized in fixed monitoring equipment,but it still has shortcomings such as limited monitoring range,blind field of vision and high cost.The technology of acquiring video images by monitoring equipment mounted on the UAV,then combining image processing to detect and track the ground vehicles came into being.However,the application of UAV in traffic monitoring and data acquisition is still immature,and reliable methods for detecting and tracking vehicles and extracting other traffic information from aerial video are still scarce.To this end,this thesis takes the vehicle on the road as the object,and uses traffic monitoring and data acquisition as the application background.The main research contents of this thesis are as follows,(1)An aerial vehicle detection algorithm based on machine learning is proposed.As the aerial view of UAV is different from the view of fixed monitoring device,four rectangular features of the vehicle conforming to the aerial view angle are added to the original Haar-Like rectangular feature database.Then the HOG feature is combined to compensate for the shortcomings of characterizing object only using Haar-Like features,a large number of positive and negative samples are used to train each feature to form a weak classifier,and finally weak classifiers are cascaded to form an Adaboost strong classifier to realize the classification and detection of the vehicle.Comparing the experimental results,the improved algorithm can effectively describe the vehicle characteristics and improve the accuracy of vehicle detection.(2)When UAV tracks the ground target,the classical Camshift algorithm is susceptible to similar color background/target,occlusion,and so on.This thesis proposes an improved target tracking method for Camshift.The tracking template based on the three-dimensional joint histogram is established by extracting the Hue,Saturation and LBP feature components of the tracking target,and the weighting values of the three feature components are adjusted by the adaptive weighting strategy to improve the tracking accuracy of the algorithm.When tracking target is occluded,the Kalman filtering mechanism is introduced to enhance the robustness of the algorithm.The experimental results show that the improved algorithm can meet the requirements of target tracking accuracy and real-time for UAV.(3)Based on the detection and tracking algorithm proposed above,a vehicle detection and tracking system based on UAV aerial photography is designed and developed in combination with UAV hardware equipment and software platform.It implements the functions of displaying road type,speed limit,vehicle driving direction,lane occupancy,average speed and vehicle count.By counting illegal vehicles that violate laws and regulations,the traffic control department is assisted to maintain traffic order and ensure traffic safety.