Research On Road Detection Technology In Autonomous Driving System

Autonomous driving is a kind of system which based on information acquisition system,computing system and decision-making system to provide unmanned vehicle solutions,it makes the modern people's life faced with a great change,it provides more security and convenient form for people's traveling,it has great significance to the formation of bringing social.The road detection technology which this paper studies is occupy the central position in the autonomous driving system,compared to the traditional road detection technology,this paper puts forward that the detection technology has not only do some optimization for road characteristics to make the lane line feature more clear,but also,it can recognizes the curve of the road as well as the multiple lanes of road according to the characteristics change in the stream of video on the process of real-time driving.The road features not clear which affected by fogging weather,the unknown lane condition and the existing corners are all directly making the autonomous driving system difficult to operate.So this paper studies the existence of the above problems in autonomous driving system,the main contents are asfollows:First of all,put forward a kind of road image defogging method,and combined with the HSV space to do the lane feature enhancement,basing on optimized segmentation threshold to achieve good lane detection.In defogging process,proposing the reconstructed depth model of fog road image,and according to value of the white lane in the S channel is 0 to do image enhancement.Experimental results showed that the proposed approach can effectively enhance image clarity;the enhancing rate of line features which have been processed through by the algorithm this paper proposed has increase by about 45%.Then,this paper uses the improved Hough transform to do the road lane line detection,and do the judgments of slope after the road lane line fitting,to reach the effect of corner detection.The experimental results show that the curves with the right driving didn't happen big change and its driving deviation rate was around the 5%.Lastly,combined with the optimization of vehicle detection,proposed a kind of assistant method of multi lane detection on the condition of vehicle exists before,the effect is significant.Using Adaboost algorithm to do road vehicle training,in the process of the actual driving,and put forward the correlation filter to do secondary testing to reduce the error detection rate.According to the relation between detected vehicle's position in front and the driver's position,to guide the judgment of the lane in the condition of the vehicle barrier exists in the running process,it can give the result of single or multiple lanes surely,and combined with vehicle control system to do vehicle lane changing instructions,deceleration,etc.The results show that the error rate of weak classifier has reduced nearly 6%,and the multilane detection direct explicitly.In conclusion,this paper's research carrying out on the road detection is effective and achieved certain effects,to a certain extent solved the detection problem of fogging road lane line,corner lane and multi lane.