Research On Vehicle Detection Technology Of Low Illumination Remote Sensing Image

In recent years,with the development of intelligent transportation system and remote sensing technology,vehicles detection technology in remote sensing images has received more and more attention.Some objective factors will have a negative impact on the image when we get the remote sensing images.The remote sensing images often have the characteristics of global or local dark when images were taken in the twilight or bad weather,it is low illumination images.In this paper,the vehicle is extracted from the remote sensing images in low illumination environment,this paper enhances the low illumination images and effectively extracts the roads and vehicle information.The main contents of this paper are as follows.At first,this paper studies the method of low illumination image processing.Low illumination image is characterized by poor contrast,the image contrast can be improved by Retinex algorithm,histogram equalization or mathematical morphology.The traditional Retinex algorithm has a halo phenomenon when the light changes drastically.Histogram equalization can improve the contrast of the image but the gray level distribution does not conform to the real scene.Mathematical morphology method can effectively improve the contrast of the image and the gray level distribution also conforms to the real scene.Next,this paper studies the road extraction technology in remote sensing images.First using top-hat transform which can enhance the bright targets in the dark background can enhance the edge of the road.According to the direction of the part of the road is a straight line,we use Hough transform method to extract the edge of the road by detecting the straight line in the binary picture.And road show bending is not in the minority,in this paper,a road detection method based on parallel lines is proposed by the parallel characteristics of the two edges of the road,which has a good effect on the road detection.Finally,two kinds of vehicle detection algorithms are proposed.We use the mathematical morphology operation for road mask picture to reduce the influence of low illumination and through the high threshold and low threshold to split the bright vehicle and dark vehicle from the road,then,we use the area constraint to filter out the part that does not conform to the vehicle characteristics.It's the first method.Detecting the edge of the road mask picture and estimating the angle of the road by road edge,then,the angle will be used to correct the inclination of the vehicle and the minimum circumscribed rectangle will be obtained,at last,vehicle information will be effectively extracted by using the shape feature of the vehicle to remove the part does not conform the vehicle.It's the second method.