Sensor Fusion Combining 3-D and 2-D for Depth Data Enhancement

Time-of-Flight (ToF) cameras are known to be cost-efficient 3-D sensing systems capable of providing full scene depth information at a high frame rate. Among many other advantages, ToF cameras are able to provide distance information regardless of the illumination conditions and with no texture dependency, which makes them very suitable for computer vision and robotic applications where reliable distance measurements are required. However, the resolution of the given depth maps is far below the resolution given by standard 2-D video cameras which, indeed, restricts the use of ToF cameras in real applications such as those for safety and surveillance. In this thesis, we therefore investigate how to enhance the resolution of ToF data and how to reduce the noise level within distance measurements. To that end, we propose to combine 2-D and ToF data using a low-level data fusion approach that enhances the low-resolution depth maps up to the same resolution as their corresponding 2-D images