An efficient and robust visual inertial odometry framework deployable on lightweight computing platforms

Aiming at the problem of performance degradation of traditional visual inertial autonomous positioning algorithms on low-computing-power embedded computing platforms, this paper designs a set of binocular visual inertial odometry frameworks based on VINS-Fusion, which is optimised for the traditional algorithms that use modules with high computational resources, such as image processing, nonlinear problem solving, etc., and takes advantage of the graphic processing unit resources of the embedded system. By rationally configuring the program running process to achieve parallelisation and acceleration, the computational speed of the module is improved, and the running performance of the algorithm on the embedded platform is improved under the premise of guaranteeing the positioning performance.