| In the face of the complex light conditions and smoke and dust occlusion characteristics of the post-disaster environment,the visual odometry method based on the visible light image is difficult to effectively locate,so this paper takes the visual odometry method in the post-disaster environment as the research object,introduces the infrared thermal imaging camera into the observation system of the robot,and integrates the inertial measurement information of the IMU to construct a visual inertial odometry method based on the direct method,providing the robot with reliable positioning and mapping ability in the post-disaster environment.The main contents of this article are as follows:(1)Lie Group and Lie Algebra are used to describe the posture state of the robot observation system,and the projection relationship between the image pixels and the three-dimensional coordinate points in the physical environment is constructed by establishing the IMU measurement model and the thermal imaging camera observation model,and the photometric changes of the environment are modeled based on the photometric error model,which further reduces the sensitivity of the system to lighting.(2)This paper proposes a strategy based on pixel gradient value and multi-layer meshing to achieve feature point selection of dynamic gradient thresholds,and proposes an algorithm flow that uses the inter-frame IMU pre-integration value of the image as the initial value,and establishes an image pyramid to efficiently solve the initial pose of the infrared image.(3)The modeling method of the state estimation problem of the robot system is analyzed,and the improved L-M algorithm is proposed to achieve efficient solution of the constraint equation;and the back-end module optimization method of the window method combining the local window and the overall window is designed to improve the positioning accuracy of the algorithm.(4)Using absolute position error as the measurement standard,the positioning accuracy of the algorithm under visible light conditions and its stability in high dynamic motion scenes are compared and verified,and the reliability of the algorithm is preliminarily verified under infrared image conditions.Through the above research,the visual inertial odometry based on infrared image is realized,which solves the problem of failure of the visual odometry based on normal light image in the post-disaster environment with complex lighting conditions and smoke and dust,and provides technical support for the stable and efficient autonomous positioning of fire rescue robots in the post-disaster environment. |