Due to the lack of scale information,the performance of monocular cameras is limited in application scenarios such as visual odometry.Existing researchers mostly address this issue through deep learning-based approaches,yet their inference speed is slow,leading to poor real-time capabilities.To solve this problem,an explicit method based on nonlinear optimization is proposed for rapid monocular depth recovery in corridor environment.The assumption of virtual camera is adopted to simplify the solution of the camera pose angles.The depth estimation problem is transformed into an optimization problem by minimizing the geometric residual.A depth plane construction method is designed to categorize the depth of space points,facilitating swift depth estimation in enclosed structural scenarios,such as corridors.Finally,the proposed method is applied in the initialization process of monocular visual odometry,so that the monocular visual odometry could obtain real scale information and improve its positioning accuracy.The experimental results show that the relative error of depth estimation of the proposed method is less than 8.4%in the corridor scene within 3 m,and can run in real time at a speed of 20 FPS on the Intel Core i5-7300HQCPU processor.
NETL
NSTL
万方数据
