A HYBRID METRIC FOR CAMERA POSE ESTIMATION IN RGB-D RECONSTRUCTION

In this paper, we propose a hybrid metric for measuring the accuracy of the estimated camera. Existing camera pose evaluation methods are mainly built upon one of the two metrics: the geometric measurement or the photometric measurement. The geometric based method inspects the relative position between source and target points, while the photometric based method utilizes the pixel intensity as assessment. The proposed hybrid metric is represented by a vector containing both the geometric measurement and the photometric measurement. The optimal transform between a pair of frames can be found by minimizing the hybrid metric between the two frames. Based on the proposed hybrid metric, we propose an outlier removal algorithm for finding the pose outliers, and a pose estimation algorithm for estimating the camera pose. Experimental results demonstrated that the proposed hybrid metric outperforms the geometric based metric and photometric based metric in terms of the accuracy of outlier identification and camera pose estimation.