Research On Vehicle Detection From High Resolution Remote Sensing Image

Recently, with the rapid development of remote sensing technology,more and more researchers are working on small objects from the remote sensing images recognition and extraction. Vehicle target detection is one of the research hot spots. Vehicle target detection in remote sensing images has important effect on the development of intelligent traffic. It can provide traffic information for traffic management department from the overall perspective and access road traffic conditions. Also it is an important supplement to the ground traffic monitoring.In this paper, we study the method of high resolution remote sensing images vehicle detection, which mainly includes the following contents:(1) A method of combining gray scale morphology and bilateral filtering is proposed to remove road traffic signs. Through the design of gray scale morphological structure elements and setting the parameters of the bilateral filtering, complete the removal of the traffic signs. The method reduces the interference caused by the traffic signs and improves the accuracy of vehicle detection.(2) For light color car leak phenomenon, an improved threshold segmentation algorithm based on vehicle detection is proposed. At first,through two value morphology operation obtain the connected domain of remote sensing images, and set the threshold to remove the non vehicle target by using the geometrical characteristic. Then use the gray histogram characteristic set threshold to remove the interference of dark car and get light colored car layer. In order to make up for some of the light colored body is not complete phenomenon, overlay the light colored car layer with the original two values of anti image, and get a relatively complete vehicle. Finally, remove the interference of traffic signs and mark out all the vehicles detected. Experiments show that the method is effective and can be used to detect light vehicles.(3) Detection of high resolution remote sensing image vehicle targets by using histogram of oriented gradient feature. First of all, the HOG features of positive and negative sample is extracted. And take it as theinput features of support vector machine and training. Then get the required classifier. Take the effect of support vector machine parameters on vehicle detection into consideration.We have to choose the parameters of support vector machine reasonably. Finally, the vehicle targets are detected by the classifier. Experiments show that the method is feasible.