Vehicle Detection System Based On Aerial Photography Of Drones

In recent years,due to the rapid increase in the quantity of vehicles and the ever-increasing demand for the flexibility of traffic monitoring,vehicle detection technology which based aerial photography has been getting more attention.At present,the vehicle detection still has some problems.For example:the shadow detection removal effect is not perfect;a single method doesn’t apply to detect all the motion states of vehicle.This thesis chooses the vehicle detection method,which is suitable for aerial photography,to design the whole system,analyzes and improves the problems in the target detection,and develops the Android mobile phone application.Firstly,depending on the different motion states of vehicles,the detection system chooses different detection methods to improve the accuracy of detection system.If there are vehicles running normally in the video,it adopts the improved method which fuses background difference method and frame difference method;if there are vehicles running slowly or remaining still,it adopts the Adaboost learning method based on the new Haar rectangle feature library.Then,this thesis researches the existing shadow detection method,and improves the Otsu threshold method:partitioning the moving object and the shadow by using adaptive threshold twice.Comparing with the original OTSU threshold method and the triple adaptive threshold method,the experimental results show that the shadow can be removed more effectively.Next,in allusion to the defects of the background difference and the frame difference,this thesis proposes an improved fusion method:fusing the background image and inter-frame difference image according to some rules.The experimental results show that the improved algorithm can accurately track the ground targets in the aerial image.After that,this paper proposes seven new rectangle features aimed at the features of aerial-vehicle.Based on the Adaboost training method,train the classifier for the original and new rectangular feature library separately.Testing two classifiers and analyzing the experimental results.Finally,based on the single-frame images or video,a real-time vehicle detection Android App is developed.The App makes the vehicle detection system play a role in real life and is used in ITS.