Being aimed at the problems that satellite navigation is vulnerable to interference,inertial navigation relies on control points,and traditional intersection measurement methods are inefficient and are difficult to meet the emergency needs,simultaneously,traditional visual matching methods are deeply affected by illumination and viewing angles,and robustness is poor at feature-sparse scenes of deep mountains and gobi deserts,this paper proposes an emergency positioning method suitable for featuresparse scenes.First,the method is to extract ridge contours by a semantic segmentation model,and perform preprocessing including edge extraction and feature point down-sampling to solve the problem ofinsufficient discriminative information in feature-sparse scenes.And then,image matching is converted into feature point matching.In combination of in-vehicle motion constraints with long-distance imaging constraints,a descriptor integrating“local + global”spatial contexts is constructed to achieve rotation-invariant properties.At the same time,stable anchor points are introduced to build a global descriptor,and matching ambiguities caused by locally similar structures are mitigated through relative displacement vectors.Finally,based on polar geometry,the fundamental matrix and essential matrix are solved.In combination of the Perspective-n-Point (PnP) algorithm with triangulation and bundle adjustment (BA),the camera pose is optimized to derive geodetic coordinates.The experiments in feature-sparse scenes such as the Qinling Mountains and Zhouwan Town,Dingbian County demonstrate that the average positioning accuracy reaches 2.5~2.8 meters by the method,surpassing classical methods by 2~3 meters.Thoughcomputation time is moderately longer than that of the other classical methods,the method still maintains real-time capability (second-level response).The measurement efficiency is higher than that of the traditional intersection.This method achieves an optimal accuracy-efficiency balance for highway emergency positioning in mid-to-long-range ridge-dominated scenes without severe perspective rotation,exhibiting substantial engineering applicability.