Research And Application Of Real-Time Camera Tracking Techniques

In applications of augmented reality, the accuracy of three-dimensional registration which directly depends on the position of the camera has a great impact on the user’s sense of immersion. With the steps of the pose prediction, feature matching and pose estimation, real-time camera tracking techniques aim to estimate the6-DOF (rotation and translation) camera parameters based on the given image sequence or video information.The parallel tracking and mapping technique (PTAM) is a popular one of the real-time camera tracking techniques, which splits the tracking and mapping tasks in camera tracking problem as two separated tasks running in two different threads. PTAM enhances the performance of camera tracking in accuracy and robustness without affecting the real-time track, and reduces the time to processing each frame of image. However, PTAM suffers from the difficulty in matching feature points under fast camera motion and lighting changes. In this paper, PTAM is improved in matching feature points.On the basis of the Binary Robust Independent Elementary Features (BRIEF), a modified feature descriptor M-BRIEF is proposed. M-BRIEF automatically adjusts the coordinates of the image patch according to the main orientation of the feature point, and presents a new method for testing definition by introducing the adaptive threshold. Compare to BRIEF, M-BRIEF performs better in recognition rate and matching time.Besides, experimental results show that M-BRIEF has a high recognition rate under the conditions of image blur, lighting changes and large viewpoint changes. By using M-BRIEF to match feature points in PTAM to track the camera, it can improve the robustness of the camera tracking under the conditions of fast camera motion exhibiting blur and lighting changes.Finally, a game system is designed on the basis of the camera pose obtained from the modified PTAM which tracks the three-dimensional feature points and obtains the position of the player wearing the camera in the scene. Then the position is used to drive the virtual3D basket to move in the game to receive different colors of eggs dropping from the sky of the game. By doing this, the player scores, and when the total score exceeds the threshold, the player will get the prize.