TY - CHAP

T1 - Denoising strategies for time-of-flight data

AU - Lenzen, Frank

AU - Kim, Kwang In

AU - Nair, Rahul

AU - Meister, Stephan

AU - Schäfer, Henrik

AU - Becker, Florian

AU - Garbe, Christoph

AU - Theobalt, Christian

N1 - Dagstuhl 2012 Seminar on Time-of-Flight Imaging and GCPR 2013 Workshop on Imaging New Modalities

PY - 2013

Y1 - 2013

N2 - When considering the task of denoising ToF data, two issues arise concerning the optimal strategy. The first one is the choice of an appropriate denoising method and its adaptation to ToF data, the second one is the issue of the optimal positioning of the denoising step within the processing pipeline between acquisition of raw data of the sensor and the final output of the depth map. Concerning the first issue, several denoising approaches specifically for ToF data have been proposed in literature, and one contribution of this chapter is to provide an overview. To tackle the second issue, we exemplarily focus on two state-of-the-art methods, the bilateral filtering and total variation (TV) denoising and discuss several alternatives of positions in the pipeline, where these methods can be applied. In our experiments, we compare and evaluate the results of each combination of method and position both qualitatively and quantitatively. It turns out, that for TV denoising the optimal position is at the very end of the pipeline. For the bilateral filter, a quantitative comparison shows that applying it to the raw data together with a subsequent median filtering provides a low error to ground truth. Qualitatively, it competes with applying the (cross-)bilateral filter to the depth data. In particular, the optimal position in general depends on the considered method. As a consequence, for any newly introduced denoising technique, finding its optimal position within the pipeline is an open issue.

AB - When considering the task of denoising ToF data, two issues arise concerning the optimal strategy. The first one is the choice of an appropriate denoising method and its adaptation to ToF data, the second one is the issue of the optimal positioning of the denoising step within the processing pipeline between acquisition of raw data of the sensor and the final output of the depth map. Concerning the first issue, several denoising approaches specifically for ToF data have been proposed in literature, and one contribution of this chapter is to provide an overview. To tackle the second issue, we exemplarily focus on two state-of-the-art methods, the bilateral filtering and total variation (TV) denoising and discuss several alternatives of positions in the pipeline, where these methods can be applied. In our experiments, we compare and evaluate the results of each combination of method and position both qualitatively and quantitatively. It turns out, that for TV denoising the optimal position is at the very end of the pipeline. For the bilateral filter, a quantitative comparison shows that applying it to the raw data together with a subsequent median filtering provides a low error to ground truth. Qualitatively, it competes with applying the (cross-)bilateral filter to the depth data. In particular, the optimal position in general depends on the considered method. As a consequence, for any newly introduced denoising technique, finding its optimal position within the pipeline is an open issue.

U2 - 10.1007/978-3-642-44964-2_2

DO - 10.1007/978-3-642-44964-2_2

M3 - Other chapter contribution

SN - 9783642449635

VL - 8200

T3 - Lecture Notes in Computer Science

SP - 25

EP - 45

BT - Time-of-flight imaging : algorithms, sensors and applications

A2 - Grzegorzek, Marcin

A2 - Theobalt, Christian

A2 - Koch, Reinhard

A2 - Kolb , Andreas

PB - Springer

ER -