A Device Pose Estimation Method Based on Monocular Visual Odometer

When the robot detects the unknown environment, it is necessary to estimate the position and posture of the device. Traditional device pose estimation methods include GPS positioning and laser detection, which are of high cost and not available in some special occasions. With the development of machine learning and 5G technology, visual positioning has gradually become the mainstream of robot navigation. Recently, visual odometry has been applied to the pose estimation of robots equipped with monocular or stereo cameras, because of its high accuracy, low cost and excellent real-time performance. In this paper, we implement a device pose estimation method based on monocular visual odometer, which can restore the self-motion of camera from a continuous sequence of images. Firstly, the ORB algorithm is applied to extracting the feature points from the input images. Secondly, we introduced the FLANN algorithm to match the extracted feature points from the adjacent frames. Finally, a pole-constraint equation is established according to the geometric relationship between the matched feature points, and the pose transition can be restored by solving this equation. Specifically, we introduce the binary particle swarm optimization (BPSO) algorithm to optimize the image processing, and analyze the calculation speed and precision of the SIFT and ORB algorithm in the same scenario. Compared with the previous work, the proposed method can achieve better real-time performance on various image datasets.