Vehicle Target Vision Detection And Geo-location System Based On UAV

With the rapid development of deep learning technology and the gradual maturity of UAV products,the vehicle target detection and geo-location system on aerial image based on UAV has far-reaching practical significance and broad application prospects.The traditional detection method based on artificial extraction feature is poor in generalization ability,low in detection accuracy and it is difficult to meet the real-time requirements.Therefore,this thesis proposes a method based on deep learning to realize real-time detection of vehicle targets.In addition,this thesis proposes a vehicle target geo-location method based on the sensor information of UAV.Furthermore,this thesis realizes the architecture design and development of the hardware and software of this system.The main contents of this paper are as follows:This thesis compares and analyzes a variety of target detection methods,focusing on the real-time target detection algorithm based on YOLOv3 and its optimization.Furthermore,this thesis proposes a bounding box optimization method based on scale prior and binarized image contour detection,which improves the vehicle detection results.This thesis briefly introduces the imaging model and calibration principle of the camera and proposes a vehicle target geo-location method based on real-time sensor data of the UAV.Aiming at the nonlinear imaging model of the camera,this thesis proposes a distance conversion method based on BP neural network for location estimation of detected targets.In addition,based on the results of target location,this thesis achieves target tracking and distinguishes moving and stationary vehicle targets.Combined with the designed algorithm,this thesis implements the architecture design and program development of a complete set of real-time vehicle target detection and geo-location system,including an Android mobile phone application and a workstation software.The experimental results show that: The vehicle target detection algorithm has a high detection accuracy and fast detection speed,which basically meets the actual needs;The proposed geo-location algorithm is quite reliable;The designed software and hardware system can work normally and has practical engineering significance.