Detection And Forensics Of Illegally Parked Vehicle Based On Quad-rotor UAV Helicopter

The traditional detection of illegally parked vehicles is usually completed by the traffic police patrol.Due to the limited traffic police,it is difficult to patrol the large parking areas efficiently.Therefore,it is very meaningful to automate the detection and forensics of illegally parked vehicles.This paper presents a method for detection and forensics of illegally parked vehicles based on quad-rotor unmanned aerial vehicle(UAV).The paper mainly includes the following aspects:First,the recognition of the vehicle is realized.The vehicle is identified from the image obtained by the UAV airborne camera.There are two main steps in vehicle identification: image feature extraction;the extracted features are input into the machine learning model,and the vehicle recognition model is trained.Gradient histogram feature(HOG)of image is selected in this paper;there are many machine learning methods,the mixture of Deformable Part Model(DPM)and Convolutional Neural Network(CNN)are used to identify the vehicle in this paper,and the two methods are compared.The average accuracy(mAP)of the CNN model is 95.42%,and the average accuracy of the DPM model is 89.56%.Therefore,the CNN model is used to identify vehicles in subsequent experiments.Second,vehicle attitude is recognized based on the shape feature of vehicle.The Speeded Up Robust Features(SURF)is used to match the feature points of the same vehicle in two images,so as to determine the location change of vehicle in the image.Third,Stroke Width Transform(SWT),K mean clustering,image morphological operation,connected-through field mark and corner extraction are used to extract parking lines from the neighborhood of vehicle images,and whether the vehicle is illegally parked is judged by detecting whether the parking line image is complete.Fourth,the airborne camera is calibrated,parameters of the camera is obtained;the position of the vehicle is calculated relative to the unmanned vehicle by the coordinate transformation and calculate the appropriate location and appropriate angle of the camera.Fifth,based on the license plate's color and texture features,the location of the license plate is accurately located from the vehicle image,and a solution is proposed for how to adjust the camera when the license plate image is incomplete or does not contain license plate image.Finally,the DJI M100 quad-rotor UAV is used as the experimental platform,the camera is mounted to the two axis gimbal,and the gimbal is mounted to the bottom of the UAV;image data is collected by the camera,vehicle detection,judgment of illegally parked illegally parked vehicles,calculation of position and attitude estimation are carried on Firefly-RK3288 development board,the board will send UAV position and attitude information relative to illegally parked vehicles to the flight control of UAV through the serial port;UAV is controlled to fly to the right position nearby head or tail of the vehicle after receiving the position information,and camera is adjusted to the right angle to shoot clear license plate image of illegally parked vehicle.