|dc.description.abstract||Range imaging cameras measure the distance to objects in the field-of-view (FoV) of the camera, these cameras enable new machine vision applications in robotics, manufacturing, and human computer interaction. Time-of-flight (ToF) range cameras operate by illuminating the scene with amplitude modulated continuous wave (AMCW) light and measuring the phase difference between the emitted and reflected modulation envelope. Currently ToF range cameras suffer from measurement errors that are highly scene dependent, and these errors limit the accuracy of the depth measurement. The major cause of measurement errors is multiple propagation paths from the light source to pixel, known as multi path interference. Multi-path interference typically arises from: inter reflections, lens flare, subsurface scattering, volumetric scattering, and translucent objects.
This thesis contributes three novel methods for resolving multi-path interference: coding in time, coding in frequency, and coding in space. Time coding is implemented by replacing the single frequency amplitude modulation with a binary sequence. Fundamental to ToF range cameras is the cross-correlation between the reflected light and a reference signal. The measured cross-correlation depends on the selection of the binary sequence. With selection of an appropriate binary sequence and using sparse deconvolution on the measured cross-correlation the multiple return path lengths and their amplitudes can be recovered. However, the minimal resolvable path length is dependent on the highest frequency in the binary sequence.
Frequency coding is implemented by taking multiple measurements at different modulation frequencies. A subset of frequency coding is operating the camera in a mode analogous to stepped frequency continuous wave (SFCW). Frequency coding uses techniques from radar to resolve multiple propagation paths. The minimal resolvable path length is dependent on the camera's modulation bandwidth and the spectrum estimation technique used to recover distance, and it is shown that SFCW can be used to measure depth of objects behind a translucent sheet, while AMCW measurements can not. Path lengths below quarter a wavelength of the highest modulation frequency are difficult to resolve.
The use of spatial coding is used to resolve diffuse multi-path interference. The original technique comes from direct and global separation in computer graphics, and it is modified to operate on the complex data produced by a ToF range camera. By illuminating the scene with a pattern the illuminated areas contain the direct return and the scattering (global return). The non-illuminated regions contain the scattering return, assuming the global component is spatially smooth. The direct and global separation with sinusoidal patterns is combining with the sinusoidal modulation signal of ToF range cameras for a closed form solution to multi-path interference in nine frames. With nine raw frames it is possible to implement direct and global separation at video frame rates. The RMSE of a corner is reduced from 0.0952 m to 0.0112 m. Direct and global separation correctly measures the depth of a diffuse corner, and resolves subsurface scattering however fails to resolve specular reflections. Finally the direct and global separation is combined with replacing the illumination and reference signals with a binary sequence. The combination allows for resolving diffuse multi-path interference present in a corner, with the sparse multi-path interference caused mixed pixels between the foreground and background. The corner is correctly measured and the number of mixed pixels is reduced by 90%.
With the development of new methods to resolve multi-path interference ToF range cameras can measure scenes with more confidence. ToF range cameras can be built into small form factors as they require a small number of parts: a pixel array, a light source and a lens. The small form factor coupled with accurate range measurements allows ToF range cameras to be embedded in cellphones and consumer electronic devices, enabling wider adoption and advantages over competing range imaging technologies.||