Research On Unstructured Road Extraction Algorithm Based On Monocular Vision

At present,the intelligent transportation and vehicle automatic driving system have become an important research direction for related industries,and the road image processing based on vision is the key technology.Compared with structured roads as urban road and highways,unstructured roads in suburb or fields have complex road environment and various imaging effects.The unstructured road image processing technology has not matured yet.Based on these,this paper studies the related technology of complex road image processing based on monocular vision.The main research contents are as follows:In the stage of road image pre-processing,this paper proposes an invariant image extraction methods based on statistical analysis to solve the shadow problem of unstructured road images.The best projection angle in logarithmic space of the image is calculated by the methods with linear regression,absolute difference information or standard deviation of the skewness.And the processing effects of proposed methods are more stable than the existing ones.For geometric information extraction of road image,this paper proposed the vanishing point detection method based on road boundary region estimation.This method selects pixels in road boundary region first,and then determines their boundary directions with multi-directional filters to take the line soft voting based on maximum weight,finally the vanishing point is detected via the focus concentration analysis method of voting images.Compared with the existing methods,the proposed method substantially improves the computational speed while ensuring the accuracy.Then,a new boundary direction extraction method based on regression statistics is proposed for pixel processing in road boundary region,and the speed of calculation is further improved.For road region extraction,this paper proposes a simple and effective road extraction method based on vanishing point detection method proposed previously.This method obtains the probability maps with some road structure reference points and the gray and edge information of the invariant image.The road region can be extracted after the probability maps integrated.Then,a method of road region extraction based on quadratic estimation is proposed.This method gets multiple road region probability maps from illumination invariant image,then the probability maps are merged and modified.After processing the probability map with the saliency optimization method based on frequency domain analysis,the geometric model of road region can be estimated first,and the final road region is extracted by combining this model and the gradient information of illumination invariant image.The methods proposed in this paper are simple in principle,accurate in road extraction,fast in calculation,and they have high stability for road images in various environments.