Research On Detection Technology Of Road Accessory Facilities Based On Aerial Image

With a burst of construction of road infrastructure,traditional road management methods and technologies become increasingly unable to meet the requirements of road management.Road management departments are now constantly exploring and pursuing more efficient management technologies and methods.At the same time,with the opening of airspace door and development of unmanned aerial vehicles,aerial images of roads and their affiliated facilities by unmanned aerial vehicles for detection can provide a new solution for road management department.Object detection based on aerial images by unmanned aerial vehicles is an important field in artificial intelligence research.Mostly through computer visual technology and image processing technology to analyse,process and understand images,the Object recognition and detection can achieve detection and classification of images.Also,applying this technology to road management department is a big progress in terms of putting into use of this technology.Thus,it is of great significance to explore aerial images by unmanned aerial vehicles regarding image recognition and detection.This dissertation makes an analysis of detection of road and traffic sign based on aerial images as per requirements of road management department.Firstly,it analyses the current research situation towards road detection and traffic sign detection as well as difficulties and existing problems in road detection and traffic sign detection.Then,it proceeds to analyses relative theories about road transport object detection,compares the merits and demerits of the segmentation methods of four images in details.It also generalizes characteristics of traffic signs and introduces description methods of color characteristics and methods of shape match regarding the characteristics of color and shape.Moreover,it makes an introduction of merits and demerits of image local features of object detection and study theory and sorting algorithms of machines.It again makes a study on road detection of aerial images in details.It proposes a kind of graphic cut of improved Graph Cuts smoothness due to the failure of traditional Graph Cuts to partition boundary very well as per image structure characteristics and also due to the neglect of different contrast ratio in different areas by aerial image of high resolution.It proves by experiments that this algorithm is of high efficiency and robustness.Then,it does a research on traffic sign detection of aerial images and puts forwards a kind of space color enhance algorithm of improved RGB regarding the degeneration brought about by the changes of cloud deck photography angle and equipment shake as well as problems such as color degeneration and shadow of traffic sign it self,which effectively enhances ROI area and restrains background area.Then threshold segmentation proves that this improved algorithm can effectively extract the target area.Following the above result,it achieves ROI through applying morphological processing and shape match;it reduces false detection and neglected detection by verification of the above ROI through machine learning methods.This dissertation applies SVM training classifier and makes experiments of the above situations through well-trained classifier.Finally,it proves that this algorithm is of good efficiency and robustness.Shortcomings are also pointed out at the end of this dissertation as well as generalization of the dissertation is made.The future research is also put forwards in this dissertation.