Low Bit-rate 3D feature descriptors for depth data from Kinect-style sensors

Abstract

In applications that require an input point cloud to be matched with a set of database point clouds present on a remote server, it is preferable to compress and transfer 3D feature descriptors online, rather than compressing and transferring the whole input point cloud. This is because the former would require much lesser bandwidth and does not require feature extraction on the server. Existing real valued 3D feature descriptors that offer good keypoint matching performance require higher bandwidth for their transfer over the network. On the other hand, the existing binary 3D feature descriptor requires relatively less bandwidth but offers reduced keypoint matching performance. In this paper, we propose to employ lattice quantization to efficiently compress 3D feature descriptors. These compressed 3D feature descriptors can be directly matched in compressed domain without any need for decompression, hence drastically reducing the memory footprint and computational requirements. We also propose double stage lattice quantization to achieve even more compression in the case of SHOT 3D feature descriptor. We provide a spectrum of possible bit rates and achievable keypoint matching performance for three state-of-the-art 3D feature descriptors. Experimental evaluation on publicly available benchmark dataset highlights that the compressed 3D feature descriptors require much lesser bandwidth and yet offer good keypoint matching performance. The source code is made publicly available for the benefit of the community.