Binocular vision localization based on vision SLAM system with multi-sensor fusion

Abstract

Simultaneous Localization and Mapping (SLAM) is a popular research topic in robotics and its key technology with very promising application scenarios. In recent years, feature point-based vision SLAM schemes have become relatively mature. However, the performance of such algorithms decreases dramatically when feature points are sparse or unevenly distributed. Vision sensors are prone to missing feature points in short-time fast motion or texture-free regions, which cannot guarantee good feature tracking and positional estimation performance. The inertial measurement unit (IMU) can well compensate for this deficiency and the vision sensor information also happens to correct for the effect of IMU drift. Visual inertial odometry is gradually becoming an important research direction in the field of SLAM for mobile robots by virtue of its high accuracy, high complementarity, and high robustness which combines data from both camera and IMU sensors. This paper deals with binocular vision localization with multi-sensor fusion in the scenario of electric power inspection. The two localization and navigation methods of vision and inertia are highly complementary and the fusion of the two can improve the accuracy of the system which is of great research value and significance.