No abstract

ToF sensing scheme Mechanical rotation of the cameraReduction of the exposure time Sparse signal recovery Sensing Matrix ConstructionSliced OMP

In this work, several ToF sensing schemes which tackle the challenge of obtaining high angular resolution and long ranges in nearly real-time, with relatively simple implementation and low associated computational load, are proposed. Sliced Orthogonal Matching Pursuit (OMP) divides the spatial domain in a number of partitions which ensure an efficient projection of the scene by optimizing the inter-column coherence of the sensing matrices. The signals are preliminarily localized within the partitions and OMP is then applied to recover depth and amplitude in a refined spatial domain. The preliminary set of measurements reduces the effective domain of the signal, lowers the processing times, and improves the sensing accuracy. Several methodologies are described for the construction of the sensing matrices, such as Low-Density Parity-Check codes via Progressive Edge Growth (LDPC-PEG), and random permutations of (0,1)-binary columns generated as combinations without repetition of a fixed number of non?zero elements for each column. Furthermore, we extend the previous schemes accounting for the raising and falling edges in order to avoid any possible coincidence which may degrade the coherence. Then, the upper super-resolution limit is studied accounting for the Instrument Response Function (IRF) of the ToF sensor. Zoned APEG extends the applicability of Adaptive Progressive Edge Algorithm (APEG) to more practical illumination systems by considering several groups of signals, arising from different areas of the sensor array, during the adaptation of the sensing matrices. The signals are then individually retrieved for each pixel via OMP over the identified joint signal support.

In this work, we propose a sensing scheme for the reconstruction of a highly sparse 3D view by a Pulse-Based Time-of-Flight (PB-ToF) camera aiming at achieving high angular resolution at interactive rates. The construction of the sensing matrices is focused on the optimization of the coherence and the preservation of the low density which characterize Low-Density Parity-Check (LDPC) codes. We investigate the possibility of shifting the custom sequences generated at the pixel level by selecting the shifts which maximize the minimum distance between adjacent columns, as well as the use of the information from the two complementary integration channels or taps the ToF sensor consists of. The signal reconstruction algorithm, coined greedy bi-lateral fusion, firstly determines a preliminary target probability distribution, and then re-weights it by exploiting the local correlations within the pixel array by applying a bi-lateral filtering which accounts for the affinities in spatial and intensity domains. The algorithm improves the accuracy of our camera under presence of strong noise and still preserves the speed and simplicity associated to classical greedy algorithms.

No abstract