Equirectangular Image Data Detection, Segmentation and Classification of Varying Sized Traffic Signs: A Comparison of Deep Learning Methods.

There are known limitations in mobile omnidirectional camera systems with an equirectangular projection in the wild, such as momentum-caused object distortion within images, partial occlusion and the effects of environmental settings. The localization, instance segmentation and classification of traffic signs from image data is of significant importance to applications such as Traffic Sign Detection and Recognition (TSDR) and Advanced Driver Assistance Systems (ADAS). Works show the efficacy of using state-of-the-art deep pixel-wise methods for this task yet rely on the input of classical landscape image data, automatic camera focus and collection in ideal weather settings, which does not accurately represent the application of technologies in the wild. We present a new processing pipeline for extracting objects within omnidirectional images in the wild, with included demonstration in a Traffic Sign Detection and Recognition (TDSR) system. We compare Mask RCNN, Cascade RCNN, and Hybrid Task Cascade (HTC) methods, while testing RsNeXt 101, Swin-S and HRNetV2p backbones, with transfer learning for localization and instance segmentation. The results from our multinomial classification experiment show that using our proposed pipeline, given that a traffic sign is detected, there is above a 95% chance that it is classified correctly between 12 classes despite the limitations mentioned. Our results on the projected images should provide a path to use omnidirectional images with image processing to enable the full surrounding awareness from one image source.

Quantitative material decomposition using linear iterative near-field phase retrieval dual-energy x-ray imaging.

This paper expands the linear iterative near-field phase retrieval (LIPR) formalism to achieve quantitative material thickness decomposition. Propagation-based phase contrast x-ray imaging with subsequent phase retrieval has been shown to improve the signal-to-noise ratio (SNR) by factors of up to hundreds compared to conventional x-ray imaging. This is a key step in biomedical imaging, where radiation exposure must be kept low without compromising the SNR. However, for a satisfactory phase retrieval from a single measurement, assumptions must be made about the object investigated. To avoid such assumptions, we use two measurements collected at the same propagation distance but at different x-ray energies. Phase retrieval is then performed by incorporating the Alvarez-Macovski (AM) model, which models the x-ray interactions as being comprised of distinct photoelectric and Compton scattering components. We present the first application of dual-energy phase retrieval with the AM model to monochromatic experimental x-ray projections at two different energies for obtaining split x-ray interactions. Our phase retrieval method allows us to separate the object investigated into the projected thicknesses of two known materials. Our phase retrieval output leads to no visible loss in spatial resolution while the SNR improves by factors of 2 to 10. This corresponds to a possible x-ray dose reduction by a factor of 4 to 100, under the Poisson noise assumption.

Linear iterative near-field phase retrieval (LIPR) for dual-energy x-ray imaging and material discrimination.

Near-field x-ray refraction (phase) contrast is unavoidable in many lab-based micro-CT imaging systems. Quantitative analysis of x-ray refraction (a.k.a. phase retrieval) is in general an under-constrained problem. Regularizing assumptions may not hold true for interesting samples; popular single-material methods are inappropriate for heterogeneous samples, leading to undesired blurring and/or over-sharpening. In this paper, we constrain and solve the phase-retrieval problem for heterogeneous objects, using the Alvarez-Macovski model for x-ray attenuation. Under this assumption we neglect Rayleigh scattering and pair production, considering only Compton scattering and the photoelectric effect. We formulate and test the resulting method to extract the material properties of density and atomic number from single-distance, dual-energy imaging of both strongly and weakly attenuating multi-material objects with polychromatic x-ray spectra. Simulation and experimental data are used to compare our proposed method with the Paganin single-material phase-retrieval algorithm, and an innovative interpretation of the data-constrained modeling phase-retrieval technique.