Performance evaluation of a Pose Estimation method based on the SwissRanger SR4000

This paper presents an experimental study on the performance of a Pose Estimation (PE) method based on a 3D time-of-flight camera - the SwissRanger SR4000. The PE method tracks the visual features in the camera's intensity image and computes the camera's pose change from the 3D data of the matched features. To attain a small PE error, the noises of the sensor's intensity and range data are analyzed and a Gaussian filter is applied to reduce the noises. The statistical property of the filtered data is then characterized and the result is used to determine the minimum number of 3D data points that are required for a satisfactory PE accuracy. Two feature extractors, the SIFT (Scale Invariant Feature Transform) and SURF (Speed Up Robust Features) extractors, are used for the PE method and their performances are compared in term of PE error and computational time. Experimental results with various combinations of rotation and translation movements demonstrate that the SIFT extractor outperforms the SURF extractor in both PE accuracy and repeatability.