A Real-Time Joint Denoising and Interpolation Algorithm for MS Xbox Kinect

This section report an experimental demo of a denoising and interpolation algorithm for the MS Xbox Kinect sensor to enhance the quality of acquired 3D models. Details are to be found on papers
  • S. Milani, G. Calvagno, "Correction and interpolation of depth maps from structured light infrared sensors", Signal Processing: Image Communication, Volume 41, February 2016, Pages 28-39, ISSN 0923-5965, http://dx.doi.org/10.1016/j.image.2015.11.008.
  • S. Milani, G. Calvagno, "Joint Denoising and Interpolation of Depth Maps for MS Kinect Sensors", Proc. of ICASSP 2012, Kyoto, Japan, Mar. 25-30, 2012.
You will find them in the publications section. Test results are encapsulated in a video presentation.


It is also possible to watch a shorter version here.

A Real-Time Low-Complexity 3D Video Cognitive Source Coder Based on MS Xbox Kinect

This section report an experimental demo for a low complexity real-time version of the CSC coder in the conference paper. Test results are encapsulated in a video presentation.



Test results for a Cognitive Source Coding of 3D Video sequences

This section report some experimental results for the Cognitive 3D Video Coding approach described in the conference paper. Test results are encapsulated in a video presentation.

Some test results for game-theory based packet classification

This section report some experimental results for the Game Theory based approach described in the conference paper 1, where a packet classification strategy based on the principles of non-collaborative games is applied for the distributed downloading of a video sequence. In the experimental setting, we assume that the uploading peers can not communicate among each other because of delays and the topology of the network. As a matter of fact, an optimal configuration can not be found since each node is not aware of the classification choices of the other peers. The proposed approach proves to be quite effective since it improves the quality of the reconstructed sequence with respect to other classification approaches. Here we report the results obtained with 3 different methods (as well as the original sequence). The whole video sequence has been coded a second time in order to reduce the size of the file to download in this operation, we have paid attention to the fact that no significant additional artifacts have been introduced by the second compression The visible artifacts are due to the prformance of the adopted classification methods, as results in the paper testify.

Some test results on rho-optimized UEP of video streams

Performance of the adaptive optimization algorithm for cross-packet FEC codes (sequence "Foreman"). The avi files contain the reconstructed sequences with rho-optimized (or act-optimized) FEC protection (on the left) compared with the sequence that is obtained without FEC protection (on the right) after error concealment. The quantization parameter is 17 for the FEC-protected sequence and 15 for the sequence without FEC protection. The lower plot reports the allocated redundancy for the cross-packet FEC channel coder Reconstructed sequences after error concealment for two different joint source-channel Rate Control (RC) approaches. The avi file report the sequence obtained from the length-adaptive joint source-channel RC algorithm (on the left) and the rho-adaptive joint source-channel RC approach (on the right). The coded sequence is "Foreman" with target bit rate 648 kbit/s (at UDP level) and target channel code rate 27 %. In the following list we report different extracts from the reconstructed sequences.


For further information, please contact me.
You can contact me writing in Italian English French Spanish.