Assistive Technology Research, The Monocular Vision-based Vehicle Driving

In recent years, vehicle safety has become an important issue for vehicle and transport system development and is a trend of future vehicle designs. A variety of surveillance sensor system used for vehicles does not only ensure a safer traffic but can also lay solid technical foundations for the autopilot system for intelligent vehicles in the future.This paper mainly focuses on lane detection and extraction, and forward vehicle detection and location methods based on monocular digital video for Intelligent Vehicles. The digital video sensing means are cheaper, more mature, and can provided rich information compared to other means. However compared with other types of sensing methods, the information provided by digital video-based is more indirect and a number of extra method are needed to extract the key information, which is quite challenging work. The lane detection and extraction is to extract the region of the brighter lane markings painted on the surface of the road from the designated area of the digital image, and retrieve the vehicle status such as location and deviation of the driving direction. The forward vehicle detection and location is based on the calibration of the camera, which links the displacement in the road plane in the real world and coordinates in the image space, in order to calculate the distance from the camera to each vehicle which is detected from the image, which provides supports to anti-collision warning systems, automatic track shifting systems and automatic navigation systems.In this paper, the main work and research results are as follows:1. To verify the double-sided links between the coordinates in the image space and the coordinates in the road planar space in the real world, we have studied on the methods for the calibration of the camera, and have determined a double-sided transform algorithm between the two coordinate spaces. Based on the camera calibration, a method to determine of the process region in the image for different practical situations is proposed.2. Based on a variety of algorithms for forward vehicle detection, and particularly the shadow region based vehicles detection algorithms, we designed an improved and a more stable forward vehicles detection and localization method.3. Under the small scale linear assumption, we present an edge line segment extraction method for extracting image edge line segments of the road markings based on the Hough transform feature extraction method for lane line segment detection; and through the analysis of the lane edge line segment extracted from the images, the vanishing point to which lane lines point can be determined. Furthermore by comparing the position of the vanishing point determined by lane line segments and the one determined by calibrated camera, much information can be drawn such as the difference between vehicle direction and road direction.4. Based on DirectShow media-streaming library by Microsoft, a camera streaming data acquisition program is developed. It has achieved a variety of ways to read the digital images collected by the camera, to meet the demands of different circumstances for video capture.