Vehicle Road Detection Algorithm Based On Binocular Vision

With the rapid development of the transportation industry,and the amount of car ownership continues to increase,traffic accidents gradually increase.The detection of traffic objects in driving scenes has become a hot topic of ITS.Accurate and efficient determination of the road surface area and traffic obstacle location will help smart vehicle systems understand the road environment.It can achieve active and safe driving,avoid possible dangers in traffic,and improve traffic safety and efficiency.In actual traffic,smart vehicle systems often obtain various types of traffic information through sensors.Among them,image vision sensors are characterized by rich content,high accuracy,low cost,and strong anti-jamming capability.Therefore,image vision sensors become an important method,which is widely used in the detection of smart cars.Detection based on stereo vision neither require prior knowledge and model construction of traffic objects,nor is it sensitive to background changes caused by weather changes such as shadows,light,and reflection.Based on this,this paper proposes a traffic object vehicle and road detection algorithm based on binocular vision.The main accomplishments of the paper include:(1)Improved road surface detection algorithm for traffic object roads to achieve road surface area detection.The disparity map and U-V disparity map are obtained by the SGM disparity estimation algorithm,and the recognition of the target is converted into a simpler line segment detection.According to the characteristics of the road itself in the binocular vision,reclassified the misclassification point,found the missing point of the circuit surface,obtained road surface area.(2)A traffic object vehicle detection algorithm is proposed to implement vehicle contour detection.On the basis of the road detection results,the region of interest is determined to segment the image,the area of the obstacle is initially locked by the U-V disparity map,the threshold area is filtered,and the symmetry analysis of the vehicle and the similarity information of the vehicle are used to obtain the vehicle contour detection results.(3)Experiments are conducted in three standard datasets(Enpeda,KITTI,Daimler).The road types include uphill sections,horizontal turn sections,and downhill sections;traffic scenes are divided into general road scenes and congestion traffic scenes containing multiple vehicles and at the same time and contains complex traffic scenes for pedestrians and vehicles.For the road surface inspection results,the accuracy rate P,the recall rate R,and the comprehensive evaluation index F are used for analysis and evaluation;for the vehicle detection results,the accuracy,recall rate,and accuracy rate evaluation index are used for evaluation.