Article Preview
TopIntroduction
At present, automatic driving and vehicle assisted driving systems have become an important research field of intelligent transportation. They provide drivers with a variety of safety assistance functions. Among these functions, lane keeping function is the key function. However, there is no lane line or even clear road boundary on unstructured roads. It is obviously impossible to keep a normal lane keeping function for automatic driving or vehicle assisted driving system on such roads. Therefore, by recognizing the road region in unstructured roads and fitting the lane lines of this road region, the reliability of automatic driving or vehicle assisted driving system on unstructured road will be greatly improved.
To solve the problem that automatic driving and vehicle assisted driving system cannot reliably work on the unstructured roads that only have fuzzy lane boundaries or even have no lane lines, and there are many interference factors around the scene. In this paper, an unstructured road segmentation method based on RGB entropy and two-dimensional maximum entropy is proposed, which jointly extract the lane region in unstructured road in advance, and then the joint determination and segmentation method of unstructured road boundary based on fuzzy entropy is applied to determine the uncertain fuzzy region in the mixed entropy image, so as to determine the lane region. The improved least square fitting quadratic curve model is applied to largely reduce the interference of interference points on lane boundary fitting, and improving the fitting speed and accuracy.
The experimental results show that our method can largely improves the ability of percepting and detecting the boundary of fuzzy road or road without lane lines under complex environment with interference. By counting the recognition time of each frame in unstructured road video that collected, the results show that our method takes less time, and the processing speed will not change with the environment and video frame rate.
In the lane recognition experiments with different vehicle speed, different collection frame rate and different lighting condition under different environment of unstructured road, our method can also ensure that the lanes in unstructured roads in each frame can be recognized quickly from low vehicle speed to high vehicle speed. This meets the real-time requirements of vehicle driving assistance system, and improves the ability of lane keeping on unstructured road for automatic driving and vehicle driving assistance system under interference environment.