Research On Intelligent Acquisition Technology Of Road Continuous Image Based On Programmable UAV

Unmanned aerial vehicle has the advantages of small size,flexible control,simple operation and easy to use and open source for secondary development.It can improve the efficiency of road performance testing and make up for the shortness of testing vehicles in the field of road non-destructive testing.At the same time,the manufacturing cost of UAVs has dropped significantly.This situation has greatly promoted the widespread use of UAVs in the road sector.Its excellent flight control capabilities and high-quality visual data collection functions can help solve the problem of insufficient detection range coverage.However,the current research based on UAV’s acquisition of road information does not consider the impact of the environment,and ignores the flight stability performance and flight parameters required during operation.Moreover,in the process of manual control,it is easy to cause the UAV to always be in a state of adjustment and fail to fly smoothly.Nowadays,the research is more focused on the use of image algorithms to identify pavement diseases,and there is a lack of research on combining the results to obtain a complete picture to locate the disease and overall evaluate the health of the road.In view of this,on the basis of the UAV’s flight control system hardware composition and flight control algorithm,simulation and flight experiments are designed to verify that the UAV can automatically and accurately perform detection operations in a relatively stable road environment.Combined with DJI MSDK and AMAP SDK,a functional program has been developed to realize the continuous image acquisition operation at equal distances on the road surface.The best camera parameters and the recommended values of flight parameters for the acquisition operation are explored.Furthermore,the MATLAB platform preprocesses the results and splices them based on SURF feature and gradual in and out fusion.Finally,the road length map obtained is verified by the edge detection operator.Firstly,the mathematical and dynamic models of the six-rotor UAV are extracted.Combined with the PID control algorithm commonly used in the industry,the sixdegree-of-freedom rotor dynamic model of the UAV flying in a stable environment is established through the MATLAB/Simulink platform.Under the specified input conditions,the drone can quickly converge to a stable flight displacement and attitude change curve within 3s.The offset is on the order of 10-1 and 10-2 respectively.In the simulator provided by DJI,the 4m/s wind speed is perpendicular to the drone flight direction and parallel to the drone flight direction.The change of position and attitude parameters under the action of wind speed load does not affect the range of collected road images.The deviation is within ±0.15 m,and the impact is small.At the same time,field flight experiments including fixed-point hovering,trajectory tracking and attitude stabilization are carried out to verify the stability and control effectiveness.The maximum trajectory deviation is 1m and the deviation of attitude and position control is within ±0.15 m.Finally,the results provide the guidance selection of working environment for road image shooting.Secondly,in order to enable the drone to accurately perform tasks in a highly automated scenario,a target application was developed.It has the basic hardware communication function to realize the connection between the mobile phone and the control system.It can manually control the UAV and add a route mode to collect the road surface.The application also integrates the Dijkstra algorithm to realize the function of automatically calculating the road in 1s after marking the start and end points of the road detection range.After that,it can autonomously controll the camera to collect road images at equidistant intervals.Finally,taking the actual in-service road as an example,the usability of the client’s automatic road calculation and isometric shooting are tested.It can successfully obtain continuous images that met the expected road.Thirdly,the best coincidence rate of continuous pictures obtained through the design of picture splicing experiment results based on the SURF algorithm is 20-25%.Combined with different road widths,camera structure parameters,and the clarity requirements of the image for subsequent machine identification of diseases,the task execution process of the UAV is calculated.The best flight altitude range is from 7.6m to 26 m.The maximum flight speed is 5~7m/s.The minimum shutter time is1/800~1/1200 s.The minimum pixel is 5.6~64 million.And the shooting distance interval is 3.87~13.25 m.It is recommended for drone operations to be equipped with a large aperture of 60 million pixels or more and a high exposure speed naked data transmission industrial camera.After calculated results being integrated into the Android client through a parameter configuration interface,the shooting task was performed with the actual road as the object,and it was verified that the road image obtained on the basis of the configuration parameters meets the requirements.Finally,for the local road images collected during the flight of the UAV,after preprocessing to eliminate the optical distortion of the image,the image is registered through the SURF feature sub-information.Then,it is calculated with the RANSAC to gain the affine matrix transformation to process the continuous image relationship.The optimize gradual in and out weight fusion was used to eliminate the uneven part of light and dark,and finally a smooth long map of road damage is obtained.Furthermore,the commonly used edge detection operator CANNY in the establishment of the disease contour model training library verifies that the above processing method has a certain availability of the disease recognition of the obtained continuous road map.On the basis that the flight controllability and stability of teh UAV meet the requirements of accurate road image acquisition,a flight control mobile client is developed by combining open source interfaces and externally-loaded maps.It can plan path and perform equidistant collection of road images.Furtheremore,the operation parameters for obtaining the best shooting results were explored and the road surface images were stitched together.Eventually,a complete set of road surface nondestructive testing systems that can be used for disease identification of road length map are obtained.The results have transformed UAVs in the field of road detection from an artificial auxiliary method to an intelligent automatic tool that can undertake the main detection tasks.This research will promote the new technology of drones to apply in in the field of the non-destructive testing and health evaluation.