Robustness analysis of 3D feature descriptors for object recognition using a time-of-flight camera

In this paper we propose to analyze characteristics of the feature descriptors in terms of robustness against typical disturbances in the context of the object recognition pipeline for depth data with intensity information. In terms of robustness the focus was on the occlusion handling, segmentation errors, sub-sampling of data as well as the presence of Gaussian noise in data. For this analyses we considered a set of real life data captured in an indoor environment using a time-of-flight sensor returning depth and intensity data. According to our test results the intensity spin estimator and the ensemble of shape functions type of feature descriptors proved to be the most suitable variant for such object recognition tasks.