MAROON: A Dataset for the Joint Characterization of Near-Field High-Resolution Radio-Frequency and Optical Depth Imaging Techniques

We devised a dataset, MAROON (Multimodal Aligned Radio and Optical frequency Object Reconstructions in the Near Field), that enables studying of different sensor modalities in direct comparison -- at the intersection between the optical and radio-frequency domain.

Four single-view depth sensors are used in our experiments, including three from the optical spectrum and one from the radio-frequency domain: 

• Stereolabs ZED X (Optical Passive Stereo)
• Intel Realsense D435i (Optical Active Stereo)
• Microsoft Azure Kinect (Near Infrared Time of Flight)
• A submodule of Rohde & Schwarz’s QAR50 (Radio-Frequency Time of Flight, MIMO Imaging Radar).

Furthermore, we measure the ground truth surface with an optical multi-view stereo (MVS) system comprising five calibrated DSLR cameras.

The dataset comprises static and quasi-static, i.e., with slow, minimal motion as in case for human hands, targets of differing materials and geometries, captured at multiple distances from all sensors simultaneously.