Vision-based Vehicle Guidance Lane Detection

F rom automatic transmissions to cruise control to antilock braking systems, passenger vehicles have incorporated automated subsystems that make driving easier and safer. Moreover, a growing number of microprocessors on late-model automobiles now monitor and control engine operations to improve mileage and reduce emissions. Thus an eventual move to autonomous, self-guided vehicles might be viewed as a logical progression rather than a radical break with the tradition of entirely human-controlled vehicle operation. The ability to perceive, or sense, the surrounding environment is essential to driving and thus to the development of autonomous self-guided vehicles. Though typically associated with living organisms, perception can also be performed by radio, acoustic, magnetic, and tactile sensors. These active sensors can measure quantities directly and generate small amounts of data; however, there are applications in which only machine vision, which acquires data noninvasively (by passive sensors such as cameras), can be successfully employed. For example, when multiple robots move about in the same indoor environment, the signals they emit may interfere with each other. This problem is greater outdoors in an unstructured environment, where many vehicles may be moving simultaneously. Hence, passive sensors are far better for automobile traffic applications. But automatic vehicle guidance in outdoor environments presents other problems intrinsic to the use of vision. For example, such environments do not present structured information. Passenger vehicles should be able to traverse all types of roads, without the integration of expensive infrastructure modifications. Without structure, key parameters such as illumination and contrast cannot be assumed but must be measured directly by the vision sensor. Hence, processing must be robust enough to adapt to different road and weather conditions and to tolerate sudden changes from sunlight to shadow—or even to the inside of a tunnel. Camera movement and drifts in its calibration must be tolerated, too. In addition, image processing for automatic vehicle guidance requires equipment that can compute a lot of data rapidly, because vehicle speed is proportional to, and bounded by, the processing rate. Computer architectures and processing techniques optimized for real-time performance offer a solution , but only low-cost systems can succeed in the marketplace. Thus, designers must use off-the-shelf components or ad hoc dedicated low-cost solutions. Despite its demands and complexity, computer vision offers a powerful way to sense the environment. It has been widely used for such vehicle-related tasks as road following, platooning (where an automatic vehicle follows a manually driven …