Study On Road Network And Automobile Information Extraction Based On High Resolution Remote Sensing Image

With the rapid development of economy and the increase of automobile quantity, road trafficproblems, such as transportation planning and management, traffic congestion and traffic accidents,have become the bottleneck which restricts urban and rural economic development of our country.Especially in recent years, natural disasters such as rainstorm, snowstorm and debris flow frequentlyoccur in some areas in China, because of changeable climate and environmental deterioration. Howto quickly understand the damage information of road and traffic state in disaster areas has been veryimportant to emergency rescue and reconstruction after disaster. Therefore, in order to fully obtainroad network and vehicle information in some regions, we need to adopt a new traffic informationcollection method which has wide cover range, good real-time performance and high accuracy, so asto solve the traffic problem and provide information for emergency rescue.This paper carried out the research on road network and vehicle information extractionalgorithm from high resolution remote sensing image. Extracted road network of different spectrabased on high resolution remote sensing image. The road network edge information is used to createa region of interest (ROI). And then this paper carried out texture feature extraction and targetrecognition based on Support Vector Machine (SVM) of ROI. Finally the vehicles were detectedsuccessfully. The main research contents included:1. Road network extraction from high resolution remote sensing images. Firstly the image wasbeen image preprocessing based on road feature, including image enhancement (contrast transform,filtering processing) and morphological processing, and the roads and non road regions were primaryseparated; then the image was carried out the edge detection, according to edge feature parameters,the road edge was extracted and non road edge was removed, and the road network edge informationwas added to the original remote sensing image.2. Sample classification based on texture features and Support Vector Machines. The samplelibrary were made up of vehicle samples and negative samples which are10×10pixels. The samplelibrary was divided into training samples and test samples. The texture features were extracted basedon gray level co-occurrence matrix. Then the SVM separator was built. The kernel function andnuclear parameters were modified based on the classifying quality of test samples. 3. Vehicle targets detection of ROI. The ROI was defined based on the road network and vehiclelocation feature of high resolution remote sensing images. The moving window detected the image ofROI, it extracted all texture features of every moving window. Based on SVM separator, the vehicletargets were detected and signed. Then the vehicle detection was realized.The road network and vehicle information detection of high resolution remote sensing imagescan provide a new decision means for intelligent transportation development. It has importantsignificance to the information degree of transport sector and the remote sensing application intransport sector. At the same time, this research also has very broad prospects in dealing with naturaldisaster and other emergencies by the government.