An Improved Method Of Real-Time Camera Pose Estimation Based On Descriptor Tracking

Most of automatic navigation technologies are developed based on simultaneous localization and mapping (SLAM), which requires to estimate the pose from the camera input image in real time. The camera pose is estimated by matching the keyframe image and camera input image. To minimize errors in the matching, we present a hybrid method using a Pyramid Lucas-Kanade optical flow and features from accelerated segment test (FAST) descriptor, where the optical flow accurately tracks the displacement of keypoints in consecutive images, and the FAST is a fast keypoint descriptor. The performance of the hybrid method is compared with a method using only the ORB descriptor matching. Due to the distance of both the grey value and position between keypoint descriptors is as far as possible from the equation of camera model, we get deal with the position and gray value of the keypoint descriptors in advance and get more precise results in different experiments. In addition, different SLAM data set with the hybrid method was tested in both indoor and outdoor environments.