Research On Unstructured Road Detection And Fitting Algorithm

In recent years,the Intelligent Vehicle Assisted Driving Systems have been widely used in structured road scenes.Since the lane lines on most roads today are clear and often straight lanes,the use of vision-based structured road detection technology can be applied to normal roads and put into use.In contrast,unstructured roads(such as soil roads in rural environments)are lack of lane markings,the degree of road bending during driving is significantly different and also influenced by external environments such as light,shadows and water stains.So we may get poor detection results.At present,the researches on unstructured road detection technology are not mature,and the detection effects in many cases are not satisfactory to real-time detection.The unstructured road detection in this paper mainly includes technical research namely road segmentation,road boundary fitting and road departure warning.In view of the shortcomings of existing unstructured road segmentation algorithms that are difficult to completely separate roads and non-road areas due to the influence of illumination and shadow,this paper proposes an improved segmentation method based on HSV color space and two-dimensional Otsu algorithm for complex scenes.The method first extracts the region of interest of the picture,excluding the interference of unrelated information such as the sky above.Then,the HSV color space conversion is performed on the region of interest,and the two-dimensional Otsu segmentation algorithm is used to process the hue map and the saturation map respectively to obtain the binary image of both.Next,according to the shape characteristics of the road in the picture,the best segmentation result is selected,and the binary segmentation result is corrected by the morphological processing and the maximum connected domain extraction method.Subsequently,a method that combined with two fitting methods to fit road boundaries is applied due to the inaccurate fitting results based on the existing fitting algorithms.The method can select the linear model fitting road boundaries using the least square method,or use the random sampling consistency algorithm to perform the quadratic curve model fitting.The least squares method is based on the global sample points to obtain the fittedlines,while the random sampling consistency algorithm can eliminate the influence of the interference points.The two fitting algorithms can achieve the fitting operation to different road environments.Finally,according to the fitted road boundary line,the ideal driving route can be obtained,and according to the angle between the route and the traveling direction of the vehicle and the lateral offset of the vehicle from the road center line,so as to determine whether the vehicle is offset during driving,and further consider whether an deviation alarm should be signaled.This paper provides a feasible auxiliary solution for vehicles traveling on unstructured roads,and can still obtain road areas quickly and accurately for roads affected by the environment such as light,shadows and water stains.