Efficient visual odometer based on inertial assistance

Abstract

The invention discloses an efficient visual odometer based on inertial assistance. The efficient visual odometer comprises a visual sensor and an inertial sensor which are fixed on a carrier, a prediction value of navigation information of the carrier is calculated by using the value of the inertial sensor, the change amount of the navigation information of the carrier is calculated according to the prediction value and is judged, the predication value is output when the change mount meets a condition, and when the change amount does not meet the condition, the predication value is corrected by using the data of the visual sensor, and the corrected value is output. The efficient visual odometer based on inertial assistance has the following advantages: the motion of the carrier is predicated by the inertial sensor, so the computational intensity of the visual odometer is reduced, and the real-time performance of the above algorithm is improved.

Description

technical field [0001] The invention relates to the technical field of visual navigation, in particular to an efficient visual odometer based on inertial assistance. Background technique [0002] Visual Odometry (VO, Visual Odometry) is the key technology of Visual Real-time Localization and Mapping (V-SLAM, Visual Simultaneous Localization and Mapping). At present, the commonly used method of navigation technology is satellite navigation, but satellite navigation is easily interfered by environments such as tall buildings, tunnels and indoors. Visual navigation is the usual way of navigating when sat nav is not available. Vision sensors can directly perceive the outside world without the assistance of external active sensors, such as GPS, UWB, etc. They are less dependent on the outside world, have high autonomy, and the cost of vision sensors is low, so they can be put into the market quickly. The V-SLAM method is the mainstream of the SLAM method and has great potential...

AI Technical Summary

Problems solved by technology

[0003] At present, most V-SLAMs require better lighting and high-texture environments. Usually, the image data obtained by the visual sensor is used to extract feature points for matching, or the pose is solved by the direct method under the assumption that the illumination is constant, and the feature points are extracted. Matching takes a lot of time and has extremely high requirements on the processor. Although the direct method takes less time than the feature point method, it still requires a certain amount of computing power, and then puts forward requirements for the computing power of the processing platform, while the embedded The computing power of the platform often cannot meet the demand, and thus cannot realize real-time computing

Images