Research On Road Information Detection In Low-altitude Images

Traffic accident rapid processing is one of the important fields of intelligent transportation development.With the rapid development of national economy and the rapid increase of vehicle ownership,more frequent traffic accidents appeared.Manual measurement is the traditional way of traffic accident measurement,which is low efficiency,and a long time measurement is likely to cause traffic jams,meanwhile,measurement results is purely recorded by hand with not so high survey precision,and can neither validate the accident scene twice.With variable type of new measuring instrument and the digital camera is applied to traffic accident scene reconnaissance,the pressure of field personnel is eased to a certain extent,but a inefficient operation mode is still existing.In this way,a new technology research on the field of traffic accidents has important theoretical value and practical significance.In view of the present hot unmanned aerial vehicle market,the unmanned aerial vehicle low-altitude image of road information detection scheme is researched in this paper,containing the target recognition of vehicles and lane lines in images,and the image calibration and redraw based on the recognition results.(1)A lane line detection algorithm based on color feature point is proposed.After analyzing the principle of SUSAN corner detection,an improved targeted corner detection algorithm is proposed,which can check object with independent obvious characteristics,and make a detection results cluster specific,so the aim of lane line detection is achieved.The experimental results show that this is a kind of high precision extraction method with better images.(2)A self-calibration method is proposed which is calibrating by width of lane lines.The algorithm can calibrate an image when lane lines are in the image road,which can convert image scale to reality scale.The results of experiment show that the accuracy of calibration method is meglio,which can satisfy the demand of follow-up measurement.(3)Analyzing the optical characteristics of vehicle target in the image,then a vehicle detection algorithm based on multi-feature fusion is studied.After the analysis of common kinds of cars,a vehicle topology model is structured,which is a kind of convenient express model.Then matching with the detection results of the vehicle,finally,it's basely achieved the aim work of image vehicle detection and vectorization.At last,the paper realize the function of redraw for road images by and large,and achieve the aim of lane line and vehicle information detection in a single unmanned aerial vehicle image.