An iterative adaptive multi-state constrained Kalman filter binocular vision/inertial mileage calculation method (NN-MSCKF) is proposed to address the problem that the existing binocular vision/inertial mileage calculation method cannot accurately capture data in real time when the rescuers are performing localization calculations in obscured space. Firstly, we analyse the tracking efficiency and real-time requirements of the rescue personnel's violent and complex movements in the occluded space, design an iterative adaptive algorithm, use window data iteration to judge the excitation and trigger the initialisation condition to construct the measurement update; secondly, we study the way of evaluating and screening the number of map points and pixel differentiation, and introduce a map point optimization mechanism to improve the real-time performance of evaluating and screening map points; finally, we build a simulation and Finally, a simulation and test platform is built to validate the algorithm. The experimental results show that the algorithm improves the real-time performance by 1s, the global accuracy by 55% and the local accuracy by 88.9% compared with the MSCKF algorithm, which verifies the effectiveness of the

method.