Research On Lane Line Recognition Technology Of Panoramic Vehicle Driving Assistance System

The recent research and applications on driving assistance systems based on machine vision are mostly concentrated on the monocular camera or binocular camera with small field of view,which have sole function and always suffer from sightless spot vision and error accumulation in the mosaicing of images.However,the refractive and reflective panoramic camera obtains the surrounding information via a single sensor at once,which can maximize the driving field of view,thus realizing multiple driving assistance functions.This paper combines machine vision with panoramic imaging technology,and do some relevant research on panoramic image expansion,lane marking and arrow markings recognition.The research results can be applied in the field of intelligent driving,which has far-reaching academic significance and considerable application prospects.An improved method for the lane marking quick recognition based on Gabor filter was proposed.First,the panoramic image was expanded into a rectangular image by the concentric circle approximation expansion method,and the expanded image was processed by Gabor filter with different phases,thus obtaining the direction interval which had the highest clarity of the lane line quickly by calculating the gray value mean.In the process of edge detection with Canny operator,only the edge points in the optimal interval were used for non-maximum suppression.Finally,double thresholding and Hough transform are used to realize the rapid recognition of lane marking.An improved method for multi classification and recognition of arrow markings based on Support Vector Machines(SVM)with adaptive partitioning and binary coding was proposed.First,the coarse detection of Harris corners is applied to the arrow markings inside the lane marking,the pseudo corners are eliminated by improved Features From Accelerated Segment Test(FAST)algorithm.According to the location of the corners with the largest two ordinates,the area to be recognized is obtained.Then the SVM classifier is trained by invariant moments features.Finally,the binary encoding is used to realize the multi classification with one SVM classifier,thus achieving the recognition of the arrow markings.In order to verify the feasibility and superiority,the test frames are captured in different weather conditions to evaluate the recognition performance of the lane marking and arrow markings recognition methods.Then the methods are tested on the captured 500 frames of road videos and compared with the existing methods.The experimental results show that the proposed lane marking recognition method and the arrow markings recognition algorithms' s recognition rate are better than 93.6% and 95.0%,respectively.Additionally,it proves strong robustness against interference from the external environment,and provides favorable guarantee for the effectiveness and stability of the driver assistance system.