Research On Gradient-Based Binocular Localization Technology Using Road Signs

With the ongoing evolution of the global economy and society,vehicle self-positioning technology has emerged as a research frontier in the domains of autonomous driving,intelligent transportation,and vehicular networking.This technology offers real-time vehicular location data for autonomous vehicles and intelligent transportation systems,thereby facilitating more accurate navigation and control,and in turn,enhancing the safety and efficacy of transportation systems while mitigating issues such as traffic accidents and congestion.However,traditional GPS positioning techniques,which has a civilian accuracy of less than ten meters,do not meet the requisite lane-level positioning demands of autonomous driving systems.Additionally,in complex urban environments such as tunnels,underground parking lots,and high-density building clusters,satellite signals are subject to obstruction,thus resulting in unreliable location determinations.In view of the aforementioned challenges,this study presents a high-precision binocular vision vehicle positioning methodology that is well-suited to complex urban environments.This technique employs binocular cameras to capture images of roads containing road signs,thereby enabling lane-level positioning of vehicles based on the road signs.This research encompasses the following components:(1)Build a simple road sign database.Most existing visual localization methods rely on complex databases with a large amount of content and calculations.This thesis designs a simple database with only numerical information,which occupies less memory,has a simple structure,is convenient for subsequent calls.In addition,it is easy to maintain,and greatly reduces database overhead.(2)Extract road sign ROI images.This thesis first uses YOLOv8 for target detection to obtain the type and rough positioning coordinates of binocular road signs.Then,the road sign ROI image is obtained by processing the rough positioning area through color threshold processing,morphological operations,and connected domain constraints.The position coordinates of the four corners of the road sign are also obtained.Finally,the binocular road sign ROI image is subjected to histogram matching and Gaussian smoothing operations to optimize the subsequent matching effect.(3)Solve disparity by road sign registration.This thesis applies optical flow commonly used for estimating object motion in front and rear frames to the calculation of the disparity of left and right frames.Based on the principle of gradient-based optical flow and the planar characteristics of road signs,this thesis improves the registration of binocular road sign ROIs by minimizing image intensity errors and obtains accurate disparity through perspective transformation matrices.The final experimental results show that the accuracy and robustness of this thesis’s algorithm are higher than those of traditional stereo matching algorithms.(4)Design and implement a vehicle self-positioning system.This thesis designs and implements the overall process of the vehicle self-positioning system,including binocular camera calibration,road sign detection,disparity calculation,and position estimation.The road sign information and vehicle self-positioning results are displayed through a human-computer interaction interface.In summary,the high-precision binocular vision vehicle self-positioning methodology presented in this study holds considerable significance for the advancement of autonomous driving and intelligent transportation systems,as it offers more reliable and precise positioning solutions.In the future,this methodology could be extended to other domains such as intelligent manufacturing and smart logistics,with expansive application prospects.